From emergingtechnetwork.publicity at gmail.com Sun Aug 2 06:00:17 2026 From: emergingtechnetwork.publicity at gmail.com (=?UTF-8?Q?Gizem_G=C3=BCltekin_Varkonyi?=) Date: Sun, 2 Aug 2026 13:00:17 +0300 Subject: Connectionists: SNAMS 2026 : The IEEE co-sponsored 13th Conference on Social Networks Analysis, Management and Security, 17 - 20 November 2026 | Barcelona, Spain Message-ID: [Apologies if you got multiple copies of this invitation] The 13th International Conference on Social Networks Analysis, Management and Security(SNAMS-2026) https://emergingtechnet.org/SNAMS2026/index.php 17 - 20 November 2026 | Barcelona, Spain (Technically Co-Sponsored by IEEE Spain Section) *SNAMS 2026 CFP* Social network analysis is concerned with the study of relationships between social entities. The recent advances in internet technologies and social media sites, such as Facebook, Twitter and LinkedIn, have created outstanding opportunities for individuals to connect, communicate or comment on issues or events of their interests. Social networks are dynamic and evolving in nature; they also involve a huge number of users. Frequently, the information related to a certain concept is distributed among several servers. This brings numerous challenges to researchers, particularly in the data mining and machine learning fields. SNAMS 2026 aims to investigate the opportunities and in all aspects of Social Networks. In addition, it seeks for novel contributions that help mitigate SNAMS challenges. That is, the objective of SNAMS 2026 is to provide the opportunity for students, scientists, engineers, and researchers to discuss and exchange new ideas, novel results and experience on all aspects of Social Networks. Researchers are encouraged to submit original research contributions in all major areas, which include, but not limited to: *SYSTEMS & INFRASTRUCTURE* Systems and algorithms for social search Infrastructure support for social networks and systems Dynamics and evolution patterns of large and complex networks Social properties in systems design Learnings from operational social networks Big Data and Social Paradigms *ALGORITHMS AND MODELS* Deep Learning and Knowledge Discovery. Measurement and analysis of social and crowdsourcing systems Benchmarking, modeling, performance and workload characterization Modeling Social Networks and behavior Management of social network data Methods for social and media analysis Information propagation and assimilation in social networks Data mining and machine learning in social systems *APPLICATIONS* Novel social applications and systems Transient OSNs (e.g. Snapchat) Special purpose OSNs (e.g., Instagram, Vine) Communities in social networks Collaboration networks New models of advertising and monetization in social networks *PRIVACY & SECURITY* Privacy and security in social systems Trust and reputations in social systems Detection, analysis, prevention of spam, phishing, and misbehavior in social systems * Submission Types:* Accepted types of submissions are including: * Full papers (8 Pages), * Short papers and PhD forum (6 Pages) * Workshop papers (6 Pages) * Posters and demos (2 Pages). All accepted papers will be published in the conference proceedings and submitted to IEEE for publication. *SNAMS 2026 Joint Workshops * The 14th workshop on Big Data and Social Networking Management and Security (BDSN 2026) The 10th Workshop on Data Science Engineering and its Applications (DSEA 2026) The 12th Workshop on Online Social Networks Technologies (OSNT 2026) The 10th Workshop on Advances in Natural Language Processing (ANLP 2026) The 9th Workshop on Sentiment Analysis and Mining of Social Networks (SAMSN 2026) The 4th Workshop on Cognitive and Neural Systems (CNS 2026) *Submissions Guidelines and Proceedings* Manuscripts should be prepared in 10-point font using the IEEE 8.5" x 11" two-column format (IEEE Templates ). All papers should be in PDF format, and submitted electronically at Paper Submission Link. A full paper must not exceed the stated length (including all figures, tables and references). Submitted papers must present original unpublished research that is not currently under review for any other conference or journal. Authors may contact the Program Chair for further information or clarification. All submissions are peer-reviewed by at least three reviewers. Accepted papers will appear in the SNAMS Proceedings, and be submitted to IEEE for inclusion. *Important Dates* ? *Paper submission deadline: August 21, 2026 * *(Extended)* ? Notification to Authors: 25 September 2026 ? Camera Ready and Registration: 10 October 2026 Please send any inquiry to the SNAMS Team at: emergingtechnetwork at gmail.com -------------- next part -------------- An HTML attachment was scrubbed... URL: From boubchir at ai.univ-paris8.fr Sun Aug 2 08:42:46 2026 From: boubchir at ai.univ-paris8.fr (Larbi Boubchir) Date: Sun, 2 Aug 2026 13:42:46 +0100 Subject: Connectionists: [CfP] The 7th International Workshop on Machine Learning for EEG Signal Processing (MLESP) In-Reply-To: References: Message-ID: _[Apologies for multiple postings]_ The 7^th international workshop on Machine Learning for EEG Signal Processing (MLESP 2026, _https://mlesp2026.sciencesconf.org/_) to be held in Dallas (USA), in hybrid mode, from 1 to 4 December 2026, in conjunction with the IEEE International Conference on Bioinformatics and Biomedicine (IEEE BIBM 2026, _https://biod.whu.edu.cn/bibm2026/_). *Overview* EEG signal processing involves the analysis and treatment of the electrical activity of the brain measured with Electroencephalography, or EEG, in order to provide useful information on which decisions can be made. The recent advances in signal processing and machine learning for EEG data processing have brought impressive progress in solving several practical and challenging problems in many areas such as healthcare, biomedicine, biomedical engineering, BCI and biometrics. The aim of this workshop is to present and discuss the recent advances in machine learning for EEG signal analysis and processing. We are inviting original research work, as well as significant work-in-progress, covering novel theories, innovative methods, and meaningful applications that can potentially lead to significant advances in EEG data analytics. This workshop is an opportunity to bring together academic and industrial scientists to discuss recent advances. *The topics of interest include but are not limited to: * - EEG signal processing and analysis - Time-frequency EEG signal analysis - Signal processing for EEG Data - EEG feature extraction and selection - Machine learning for EEG signal processing - EEG classification and Hierarchical clustering - EEG abnormalities detection (e.g. Epileptic seizure, Alzheimer's disease, etc.) - Machine learning in EEG Big Data - Deep Learning for EEG Big Data - Neural Rehabilitation Engineering - Brain-Computer Interface - Neurofeedback - EEG-based Biometrics - Related applications *Important Dates * Sept. 27, 2026?(11:59 pm CST):Due date for full workshop papers submission Oct. 18, 2026: Notification of paper acceptance to authors Nov. 8, 2026: Camera-ready of accepted papers Dec 1-4, 2026: Workshops *Paper submission * - Please submit a full-length paper (up to 8 pages IEEE 2-column format) through the online submission system. You can download the format instruction here: https://www.ieee.org/conferences/publishing/templates - Electronic submissions in PDF format are required. *Online submission* https://wi-lab.com/cyberchair/2026/bibm26/scripts/submit.php?subarea=S02&undisplay_detail=1&wh=/cyberchair/2026/bibm26/scripts/ws_submit.php ** *Publication* All accepted papers will be published in the BIBM proceedings and IEEE Xplore Digital Library. *Contact* Prof. Larbi Boubchir /(Workshop Chair/) University of Paris 8, France E-mail: larbi.boubchir at univ-paris8.fr -------------- next part -------------- An HTML attachment was scrubbed... URL: From caspar.schwiedrzik at googlemail.com Sun Aug 2 15:17:56 2026 From: caspar.schwiedrzik at googlemail.com (Caspar M. Schwiedrzik) Date: Sun, 2 Aug 2026 21:17:56 +0200 Subject: Connectionists: Journal of Vision Special Issue: New Directions in Perceptual Learning Message-ID: Dear colleagues, We are pleased to announce a Journal of Vision Special Issue on New Directions in Perceptual Learning. We welcome empirical papers presenting new results from human participants and animal models, computational modeling papers, and theoretical or review articles. Submissions may focus on visual perceptual learning, as well as cross-modal and multisensory studies. Topics include behavioral and psychophysical studies, neural mechanisms, specificity and transfer, computational models, links to attention, reward, prediction, memory, and decision-making, developmental and clinical applications, individual differences, biological and artificial systems, and new methods or datasets for studying perceptual plasticity. The Feature Editors are Stephen Adamo, Marisa Carrasco, Miguel Eckstein, Krystel Huxlin, Zhong-Lin Lu, Aaron Seitz, Michael Webster, and Caspar M. Schwiedrzik. Submissions are accepted through March 15, 2027. The call for papers is available here: https://jov.arvojournals.org/ss/newdirectionsinperceptuallearning.aspx We warmly invite submissions and would be grateful if you could share the call with colleagues, students, postdocs, and others who may be interested. With best wishes, Caspar M. Schwiedrzik on behalf of the Feature Editors From smzhouy at gmail.com Sun Aug 2 07:39:22 2026 From: smzhouy at gmail.com (Shangming Zhou) Date: Sun, 2 Aug 2026 12:39:22 +0100 Subject: Connectionists: =?utf-8?q?Fwd=3A_FW=3A_Invitation=3A_Agentic_AI_M?= =?utf-8?q?asterclass_=7C_University_of_Essex_=7C_1=E2=80=932_Octob?= =?utf-8?q?er_2026?= In-Reply-To: References: Message-ID: Dear All, You may find this event interesting to you and your group. The University of Essex, in partnership with Sensiwise AI and KAUST Academy (one of the world?s top 20 research institutions) is hosting a two-day Agentic AI Masterclass on campus in Colchester this 1-2 October 2026. This is not a lecture series. Over two days you will build real, working AI agent systems from scratch, guided by world-class experts and leaving with a University of Essex certificate of completion. *WHAT YOU WILL BUILD OVER TWO DAYS* ? A RAG pipeline: ground AI responses with your own real data ? A data analyst agent: that decides what analysis to run, autonomously ? A meeting summarizer: extracting structured action items from transcripts ? A research agent: that retrieves, evaluates, and synthesises information ? Your own capstone solution: a business-ready agentic AI system, built and presented by you *EVENT DETAILS* ? Date: 1?2 October 2026 ? Time: 9:00 AM ? 5:00 PM each day ? Location: University of Essex, Colchester, CO4 3SQ ? Fee: ?650 per delegate (VAT exempt) ? Capacity: Strictly limited to 50 participants ? Certificate: University of Essex certificate of completion ? Read more: https://www.essex.ac.uk/Events/2026/10/01/Agentic-AI-workshop-2026 ? Register: https://www1.essex.ac.uk/online_shop/conference/default.aspx?conference=agentic-ai-workshop-2026-a-twoday-industry-masterclass *WHAT IS INCLUDED* ? Both workshop days with expert-led sessions and supervised labs ? Pre-configured cloud environments, no installation required ? All course materials, lab notebooks, and real-world datasets ? Catering across both days (lunch and refreshments) ? Participant handbook and goody bag ? On-campus B&B accommodation ? University of Essex certificate of completion *WHO SHOULD ATTEND* ? AI engineers and data scientists ready to build production-grade agents ? Software developers moving beyond LLM prompting into autonomous systems ? Technical professionals and researchers exploring applied agentic AI ? Academic staff and postgraduate students in AI and related fields No prior experience with agentic AI is required. A laptop with a modern browser is all you need, all lab environments are pre-configured in the cloud. *REGISTER NOW* Places are strictly limited to 50 participants and will be allocated on a first-come, first-served basis. ? Register here: https://www1.essex.ac.uk/online_shop/conference/default.aspx?conference=agentic-ai-workshop-2026-a-twoday-industry-masterclass ? Enquiries: h.raza at essex.ac.uk ? More info: hello at sensiwise.ai If this is not relevant to you personally, please do feel free to forward it to a colleague who may benefit. I look forward to welcoming you to Essex in October. -- Best Regards! *Dr Haider Raza*, PhD, FHEA, SMIEEE Senior Lecturer in Artificial Intelligence, || AI Upskilling Manager || Study Abroad Officer|| IADS: AI for Public Services Lead and Alumni Representative, School of Computer Science and Electronic Engineering, University of Essex T +44 (0)1206 87-4831 <+441206874831> E h.raza at essex.ac.uk *OW* https://www.essex.ac.uk/people/razah72409 *PW* http://sagihaider.com/ ------------------------------ This email and any files with it are confidential and intended solely for the use of the recipient to whom it is addressed. If you are not the intended recipient then copying, distribution or other use of the information contained is strictly prohibited and you should not rely on it. If you have received this email in error please let the sender know immediately and delete it from your system(s). Internet emails are not necessarily secure. While we take every care, University of Plymouth accepts no responsibility for viruses and it is your responsibility to scan emails and their attachments. University of Plymouth does not accept responsibility for any changes made after it was sent. Nothing in this email or its attachments constitutes an order for goods or services unless accompanied by an official order form. -------------- next part -------------- An HTML attachment was scrubbed... URL: From anna.deichler at gmail.com Mon Aug 3 07:13:07 2026 From: anna.deichler at gmail.com (Anna Deichler) Date: Mon, 3 Aug 2026 13:13:07 +0200 Subject: Connectionists: [HSI @ ECCV 2026] Present your ECCV conference paper at the HSI Workshop Message-ID: Dear colleagues, If you have a paper accepted at the ECCV 2026 main conference related to human-scene interaction, you are welcome to present it at our workshop as a non-archival poster, a chance to discuss your work with the HSI community. No new submission or review is required, and the paper will not appear in the workshop proceedings. Sign up by *August 24, 2026*: https://forms.gle/BLo35pVx9oAoKF6e7 Workshop: https://www.hsi-workshop.com/ Contact: hsi-workshop at googlegroups.com The HSI Workshop Organizers -------------- next part -------------- An HTML attachment was scrubbed... URL: From Yves.Grandvalet at utc.fr Mon Aug 3 09:10:27 2026 From: Yves.Grandvalet at utc.fr (Yves) Date: Mon, 3 Aug 2026 15:10:27 +0200 Subject: Connectionists: 2 lecturer positions Message-ID: Dear all, The Sorbonne Cluster for Artificial Intelligence (SCAI) is opening two lecturer positions (2?3 years, with possible extensions) at the university of technology of Compi?gne. Applications are accepted until September 30, 2026 (inclusive). Both positions focus on artificial intelligence and its applications: * AI for Industry and Risk Management (https://utc.recruitee.com/o/enseignant-chercheur-contractuel-fh-ia-pour-lindustrie?lang=en) * AI for Robotics (https://utc.recruitee.com/o/enseignant-chercheur-contractuel-fh-ia-pour-la-robotique?utm_source=cojob.fr&lang=en) The positions are open to a wide range of profiles, from theoretical to applied research, with potential topics including planning, uncertainty handling, explainability, and more. For further details, please contact the persons listed in the advertisements. With a light teaching load, these positions provide an ideal environment for preparing competitive applications to the CNRS. Best regards, Yves -- _______________________________________________________ Yves Grandvalet, Directeur de recherche CNRS, Heudiasyc UMR 7253, UTC, G?nie Informatique, bureau 219 e-mail : Yves.Grandvalet at utc.fr tel : 33(0)3 44 23 49 28 fax : 33(0)3 44 23 44 77 web : http://www.hds.utc.fr/~grandval/ address: 57 avenue de Landshut, CS 60319, 60203 Compi?gne CEDEX, France _______________________________________________________ -------------- next part -------------- A non-text attachment was scrubbed... Name: smime.p7s Type: application/pkcs7-signature Size: 1367 bytes Desc: not available URL: From terry at snl.salk.edu Sat Aug 1 23:02:09 2026 From: terry at snl.salk.edu (Terry Sejnowski) Date: Sat, 1 Aug 2026 20:02:09 -0700 Subject: Connectionists: A chief suspect for schizophrenia Message-ID: Schizophrenia is a devastating thought disorder for which we do not have effective treatments.Brain disorders with specific failure modes may give us insights into the normal function of specific brain circuits.For example, loss of dopamine-producing neurons in the midbraincauses Parkinson's disease:Tremors, muscle rigidity, slow movements, and impaired balance.If we could trace the origin of schizophrenia to? a specific type of neuron or neural circuit it might be possible to design more effective treatments. Converging evidence from experiments, autopsies and computational models have uncovered a chief suspect. https://terrysejnowski.substack.com/p/part-20-schizophrenia Terry ------ -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: pnF0qzUk7RutbV02.png Type: image/png Size: 2297933 bytes Desc: not available URL: From aidhwp1 at gmail.com Tue Aug 4 02:06:09 2026 From: aidhwp1 at gmail.com (AI-DHWP) Date: Tue, 4 Aug 2026 15:06:09 +0900 Subject: Connectionists: =?utf-8?q?=E3=80=90CFP_=28Submission_Deadline=3A_?= =?utf-8?q?Aug=2E_15=2C_2026=29=E3=80=91The_2nd_International_Works?= =?utf-8?q?hop_on_AI-empowered_Digital_Health_and_Well-being_Promot?= =?utf-8?q?ion_=28AI-DHWP_2026=29=2C_November_9-13=2C_2026=2C_Melbo?= =?utf-8?q?urne=2C_Australia?= Message-ID: [Our apologies if you receive multiple copies of this CFP] Dear Colleagues, We are writing to invite you to submit your papers at The 2nd International Workshop on AI-empowered Digital Health and Well-being Promotion (AI-DHWP 2026), in conjunction with IEEE CyberSciTech 2026 scheduled on November 9-13, 2026, in Melbourne, Australia. This workshop serves as a platform for sharing advances in AI-empowered and human-centered approaches that apply to healthcare and human behavior. We look forward to your submissions! Website: https://aidhwp.github.io/2026.html Submission Link: https://edas.info/N35527 INTRODUCTION ??????????? With the spread of information technology in healthcare, there is growing interest in using AI to support health management and well-being in daily life. As more data become available from wearables, sensors, and online platforms, new possibilities emerge for understanding health conditions, human behaviors, and predicting health risks. This workshop focuses on AI-empowered and human-centered approaches that apply to healthcare. It aims to promote digital health technologies that provide opportunities to support healthcare. We hope to bring together researchers to explore how AI can contribute to health and well-being promotion. Important Dates ??????????? - Paper Submission Due: Aug. 15, 2026 - Author Notification: Sep. 01, 2026 - Paper Registration: Sep. 24, 2026 - Camera-ready Submission Due: Oct. 01, 2026 Topics of interest include, but are not limited to: ??????????? - Big Data Analytics in Health - Causal Discovery and Inference in Health - Multimodal Data Fusion in Health - Health Analysis with Wearable Sensor & IoT Data - Health Analysis with Social Media Data - Biometric Analysis (gait, eye-tracking, falls, etc.) - AI for Human Behavior Modeling - AI for Aging and Gerontechnology - AI for Digital Mental Health - AI Agents for Digital Health - Explainable AI for Digital Health - Machine Learning and Deep Learning for Healthcare - Data-Driven Smart Health Applications - Human-AI Interaction for Health and Well-being Promotion - Metaverse and Digital Twin for Healthcare - Blockchain and Federated Learning for Digital Health - Privacy Protection Concern in Digital Health - Ethics, Security, and Fairness in Digital Health Submission and Publication ??????????? The submitted papers should be 4-6 pages long including figures and references and prepared in IEEE CS Proceedings format. We also welcome Position Statement Papers, which present novel ideas, hypotheses, and emerging research directions in AI for digital health. All submitted papers will be peer-reviewed by two or three experts in the field based on originality, relevance, clarity, and technical quality. Please follow the guideline in IEEE CyberSciTech 2026 Paper Submissions Site to submit your work via EDAS (https://edas.info/N35527). All accepted papers will be published in IEEE CPS Proceedings (IEEE-DL and EI indexed). -------------- next part -------------- An HTML attachment was scrubbed... URL: From terry at salk.edu Tue Aug 4 09:46:24 2026 From: terry at salk.edu (Terry Sejnowski) Date: Tue, 04 Aug 2026 06:46:24 -0700 Subject: Connectionists: NEURAL COMPUTATION - August 1, 2026 In-Reply-To: Message-ID: NEURAL COMPUTATION - Volume 38, Number 8 - August 1, 2026 http://www.mitpressjournals.org/toc/neco/38/8 http://cognet.mit.edu/content/neural-computation Articles Cross-Frequency Coupling as a Neural Substrate for Prediction Error Evaluation: A Laminar Neural Mass Modeling Approach Giulio Ruffini, Edmundo Lopez Sola, Raul de Palma Aristides, Roser Sanchez-Todo, Jakub Vohryzek, Francesca Castaldo, and Karl Friston A Model-free Reinforcement Learning Implementation of Decision Making Under Uncertainty by Sequential Sampling Jamal Esmaily Sadrabadi, Rani Moran, Yasser Roudi, and Bahador Bahrami Letters DROP: Distributional and Regular Optimism and Pessimism for Reinforcement Learning Taisuke Kobayashi W-Kernel and Its Principal Space for Frequentist Evaluation of Bayesian Estimators Yukito Iba A Hidden Markov Model-Inspired Sequence Classification Method for Hyperdimensional Computing Krzysztof ?lot, Jakub Bednarski, Kacper Kubicki, and Piotr ?uczak ----- ON-LINE -- http://www.mitpressjournals.org/neco MIT Press Journals, One Rogers Street, Cambridge, MA 02142-1209 Tel: (617) 253-2889 FAX: (617) 577-1545 journals-cs at mit.edu ----- From dhansel0 at gmail.com Tue Aug 4 11:26:20 2026 From: dhansel0 at gmail.com (David Hansel) Date: Tue, 4 Aug 2026 18:26:20 +0300 Subject: Connectionists: Fwd: New on the VVTNS YouTube channel : Equilibrium Geometry and Chaotic Dynamics in Large Recurrent Neural Networks | Giancarlo La Camera , Stony Brook University In-Reply-To: References: Message-ID: [image: VVTNS.png] > > > https://www.wwtns.online > > - on twitter: wwtns at TheoreticalWide and on YouTube > New on the VVTNS Website and YouTube Channel https://www.wwtns.online Equilibrium Geometry and Chaotic Dynamics in Large Recurrent Neural Networks lecture delivered on May 27, 2026 by Giancarlo La Camera Stony Brook University Abstract: Large recurrent networks are important models in several fields, including neuroscience, machine learning, physics, and applied mathematics. Yet their dynamics are difficult to study directly, because high-dimensional nonlinear systems can exhibit rich behavior that is hard to summarize in terms of individual trajectories. In this talk, I will discuss an approach that seeks to understand such dynamics through the structure of the network?s equilibria. I will focus on a random balanced network of threshold-linear units that undergoes a transition from a single stable equilibrium to extensive chaos as the disorder strength crosses a critical value. Using a combination of Kac?Rice theory, replica calculations, numerical root-finding, and dynamical mean-field theory, we show that the chaotic regime contains an exponentially large number of equilibria. These equilibria are all saddles, but with only a fractionally small number of unstable directions. Surprisingly, despite the completely random connectivity, the equilibria are not scattered randomly through phase space. Instead, they are strongly correlated and confined to a comparatively small region. The chaotic attractor lies within this same region, suggesting a direct geometric link between the organization of unstable equilibria and the collective structure of the dynamics. This picture helps explain why networks with extensive chaos can nevertheless display dynamics dominated by a relatively small number of collective modes. More broadly, the results suggest that the geometry of equilibria provides a useful complementary perspective to dynamical mean-field theory for understanding high-dimensional neural dynamics. . *About VVTNS : * *VVTNS is a weekly digital seminar on Zoom targeting the theoretical neuroscience community. Launched as the World Wide Neuroscience Seminar (WWTNS) in November 2020 and renamed in homage to Carl van Vreeswijk in Memoriam (April 20, 2022), its aim is to be a platform to exchange ideas among theoreticians, but not only. Speakers have the occasion to talk about theoretical aspects of their work which cannot be discussed in a setting where the majority of the audience consists of experimentalists. The seminars are 45 min long followed by a discussion and are held on Wednesdays at 11 am ET. The talks are recorded with authorization of the speaker and are available to everybody on our YouTube channel.* ? -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: VVTNS.png Type: image/png Size: 41084 bytes Desc: not available URL: From swickbrennan at yahoo.com Wed Aug 5 08:31:18 2026 From: swickbrennan at yahoo.com (Brennan Swick) Date: Wed, 5 Aug 2026 12:31:18 +0000 (UTC) Subject: Connectionists: [CFP][Meetings] NeuRo-SymBolic World Models (RoBoWoMo) Workshop @ IROS 2026 References: <1806120909.5421790.1785933078601.ref@mail.yahoo.com> Message-ID: <1806120909.5421790.1785933078601@mail.yahoo.com> TL;DR: Event: NeuRo-SymBolic World Models (RoBoWoMo) Workshop at IROS 2026 (Pittsburgh, US) Date: September 27, 2026 Website: https://worldmodelworkshop.github.io/ Late Submission Deadline: August 25, 2026 OpenReview submission link: https://bit.ly/SubmitToRoBoWoMo? Workshop Overview:? World models have emerged as a central paradigm in robot learning, planning, and reasoning, yet the term spans very different meanings across communities. High-dimensional neural models often suffer from data inefficiency, while symbolic systems require extensive domain engineering. This workshop focuses on bridging the gap between these isolated communities. We aim to bring together researchers across neural, symbolic, and hybrid backgrounds to clarify terminology, align assumptions, and identify shared challenges. By unifying these paradigms, we hope to elicit hybrid methods that drastically improve generalization, interpretability, and long-horizon reasoning for complex robotic domains. Call for Papers:? - Half Papers: Up to 4 pages (excluding references/appendices) - Full Papers: Up to 8 pages (excluding references/appendices) We welcome submissions covering neuro-symbolic unification, hybrid architectures, task-driven world modeling, benchmarking/evaluation, etc. All accepted papers will have the opportunity to be presented via lightning talks and a dedicated poster session, with the strongest submissions selected for longer spotlight presentations. For more details on formatting, submission policies, and topics of interest, please visit our Call for Papers page: https://worldmodelworkshop.github.io/call_for_papers/? Important Dates (AoE): - Early Submission Deadline: August 15, 2026 - Late Submission Deadline: August 25, 2026 - Late Author Notification: September 10, 2026 - Workshop Date: September 27, 2026 For schedule details: https://worldmodelworkshop.github.io/schedule? Awards: To recognize outstanding contributions and support early-career talent, we will be offering four Best Paper awards, with spotlight presentations! Invited Speakers and Panelists:? - Sherry Yang (New York University & Google DeepMind) - Yilun Du (Harvard University) - Tom Silver & Yixuan Huang (Princeton University) - Siddharth Srivastava (Arizona State University) - Emre Ugur (Bogazici University) - Sungjin Ahn (KAIST & New York University) - Jiajun Wu (Stanford University) Organizers: - Ahmed Jaafar* (Air Force Research Laboratory - AV) - Brennan Swick* (Air Force Research Laboratory - National Research Council) - Alper Ahmetoglu (Brown University) - Naman Shah (Allen Institute for Artificial Intelligence) - Josh Roy (Princeton University) - David Paulius (University of Notre Dame) - Sudarshan Harithas (Brown University) - Nathan Hertlein? (Air Force Research Laboratory) Advisory Board: - Thao Nguyen (Haverford College) - James Hardin (Air Force Research Laboratory) - Yuke Zhu (University of Texas at Austin) - Justus Piater (University of Innsbruck) - Hao Su (Fudan University) We look forward to your submissions and to seeing you in Pittsburgh! Best regards, The RoBoWoMo Organizing Committee Yahoo Mail: Search, Organize, Conquer -------------- next part -------------- An HTML attachment was scrubbed... URL: From mattnassar at gmail.com Wed Aug 5 09:01:09 2026 From: mattnassar at gmail.com (Matt Nassar) Date: Wed, 5 Aug 2026 09:01:09 -0400 Subject: Connectionists: Save the dates: Reinforcement Learning and Decision Making 2027 Message-ID: ====================================================== The 7th Multidisciplinary Conference on Reinforcement Learning and Decision Making (RLDM2027) www.rldm.org July 6-9, 2027 at Maison de la Chimie, Paris, France ====================================================== Over the last few decades, reinforcement learning and decision making have been the focus of an incredible wealth of research spanning a wide variety of fields including psychology, artificial intelligence, machine learning, operations research, control theory, animal and human neuroscience, economics and ethology. Key to many developments in the field has been interdisciplinary sharing of ideas and findings. The goal of RLDM is to provide a platform for communication among all researchers interested in "learning and decision making over time to achieve a goal". The meeting is characterized by the multidisciplinarity of the presenters and attendees, with cross-disciplinary conversations and teaching and learning being central objectives along with the dissemination of novel theoretical and experimental results. The main meeting will be single-track, consisting of a mixture of invited and contributed talks, tutorials, and poster sessions. RLDM 2027 will include a half-day of thematically-focused contributed Workshops, which will be run in parallel. To further facilitate cross-disciplinary connections, there will be multiple social and networking opportunities. Important dates and speakers will be announced in a subsequent email. Best, RLDM2027 ORGANIZERS GENERAL CHAIR Anne GE Collins, University of California Berkeley PROGRAM CHAIRS David Abel, Google DeepMind Matt Nassar, Brown University LOCAL CHAIRS Silvia Tulli, ISIR Polytech, Sorbonne Universit? Valentin Wyart, ?cole Normale Sup?rieure de Paris REVIEW CHAIRS ?zg?r ?im?ek, University of Bath Bob Wilson, Georgia Institute of Technology WORKSHOP CHAIRS Becket Ebitz, Universit? de Montr?al Giorgia Ramponi, University of Zurich, ETH & Chalmers University of Technology TUTORIAL CHAIRS Angela Langdon, National Institute of Mental Health Ian Osband, Google DeepMind MENTORING CHAIRS Lilian Weber, Osnabr?ck University Benjamin Eysenbach, Princeton University FUNDRAISING CHAIR Giuseppe Paolo, Cognizant AI Lab PUBLICITY CHAIR Elisa Massi, ETIS Lab, CY Cergy Paris Universit? TRAVEL AWARD CHAIR Heike Stein, ISIR, Sorbonne Universit? EXECUTIVE COMMITTEE Emma Brunskill, Stanford University Roshan Cools, Radboud University Nijmegen Will Dabney, Google DeepMind Nathaniel Daw, Princeton University Cate Hartley, New York University Peter Stone, The University of Texas at Austin and Sony -------------- next part -------------- An HTML attachment was scrubbed... URL: From khamassi at isir.upmc.fr Thu Aug 6 05:57:59 2026 From: khamassi at isir.upmc.fr (Mehdi Khamassi) Date: Thu, 6 Aug 2026 11:57:59 +0200 Subject: Connectionists: =?utf-8?q?Deadline_Extended_to_August_12_?= =?utf-8?q?=E2=80=93_Call_for_Extended_Abstracts_=E2=80=93_WWIN_2026_Works?= =?utf-8?q?hop_=40_SAB_2026?= Message-ID: <0947972e-f9f3-467c-9d0a-c7069511e121@isir.upmc.fr> (Apologies for cross-posting) Dear colleagues, We are pleased to announce that the submission deadline for the first international workshop *?Wanting What is Needed: From Homeostatic Control to Adaptive Behavior?*has been extended to *August 12, 2026*. The workshop will be held on October 19 in Berlin, Germany, as part of The 18th International Conference on the Simulation of Adaptive Behavior (SAB 2026). Workshop website:https://wwin-workshop.github.io/sab2026 Keywords (including but not limited to): internal states, homeostasis, allostasis, reinforcement learning, decision making, robotics, adaptive behavior ====================================== Aims of the workshop ====================================== Classical reinforcement learning and cognitive models often begin with externally specified rewards. Biological organisms, by contrast, are autonomous learning agents whose behavior is constrained by the need to sustain their own viability. This workshop asks how internal physiological regulation can ground decision-making, value formation, and motivated behavior. We aim to foster a shared interdisciplinary conversation across machine learning, artificial intelligence, robotics, neuroscience, biology, physiology, ethology, and control theory. The goal is to clarify how homeostatic and allostatic regulation can contribute to adaptive behavior in both biological organisms and artificial agents. ====================================== Invited speakers ====================================== * Lilian A. Weber (Osnabr?ck University) * Lola Ca?amero (CY Cergy Paris University) * Antonio Damasio (University of Southern California, online) ====================================== Extended abstracts for lightning talks and posters ====================================== * Extended abstracts of up to 3 pages (main text) * References, appendices, and acknowledgements are not included in this limit * Springer LNCS style (template available on the workshop website) ====================================== Important dates ====================================== * Extended Abstract Submission Deadline: August 12, 2026 * Notification of Acceptance: August 19, 2026 * Workshop Date: October 19, 2026 (afternoon) We welcome contributions from researchers working across the biological and artificial sciences and look forward to receiving your submissions. Best regards, Mehdi Khamassi CNRS / Sorbonne Universit? On behalf of the organizing committee: Naoto Yoshida ? Kyoto University Ismael Freire ? Sorbonne Universit? Boris Gutkin ? ?cole Normale Sup?rieure Henning Sprekeler ? Technical University of Berlin Louis L?Haridon ? ENSEA / CNRS Lola Ca?amero ? CY Cergy Paris University Mehdi Khamassi ? CNRS / Sorbonne Universit? -- Mehdi Khamassi, PhD, HDR Research director (DR2), Centre National de la Recherche Scientifique Institute of Intelligent Systems and Robotics, Sorbonne Universit? ISIR - BC 173, 4 place Jussieu, 75005 Paris, France phone: +33 6 50 76 44 92 email:mehdi.khamassi at sorbonne-universite.fr https://pages2.isir.upmc.fr/mkhamassi/ Co-director of the master's program in cognitive science (https://cog-sup.fr) Sorbonne Universit? / Universit? Paris Cit? Visiting Researcher at the Institute of Communication and Computer Systems National Technical University of Athens, Greece https://robotics.ntua.gr/people/ -------------- next part -------------- An HTML attachment was scrubbed... URL: From yossy0157 at gmail.com Thu Aug 6 02:18:41 2026 From: yossy0157 at gmail.com (Yoshida Naoto) Date: Thu, 6 Aug 2026 15:18:41 +0900 Subject: Connectionists: [Deadline Exntended] WWIN 2026 Workshop @ SAB 2026 Message-ID: (Apologies for cross-posting.) Dear colleagues, We are pleased to announce that the submission deadline for the first international workshop ?Wanting What is Needed: From Homeostatic Control to Adaptive Behavior? has been extended to August 12, 2026. The workshop will be held on October 19 in Berlin, Germany, as part of The 18th International Conference on the Simulation of Adaptive Behavior (SAB 2026). Workshop website: wwin-workshop.github.io/sab2026 Keywords (including but not limited to): internal states, homeostasis, allostasis, reinforcement learning, decision making, robotics, adaptive behavior ====================================== Aims of the workshop ====================================== Classical reinforcement learning and cognitive models often begin with externally specified rewards. Biological organisms, by contrast, are autonomous learning agents whose behavior is constrained by the need to sustain their own viability. This workshop asks how internal physiological regulation can ground decision-making, value formation, and motivated behavior. We aim to foster a shared interdisciplinary conversation across machine learning, artificial intelligence, robotics, neuroscience, biology, physiology, ethology, and control theory. The goal is to clarify how homeostatic and allostatic regulation can contribute to adaptive behavior in both biological organisms and artificial agents. ====================================== Invited speakers ====================================== * Lilian A. Weber (Osnabr?ck University) * Lola Ca?amero (CY Cergy Paris University) * Antonio Damasio (University of Southern California, online) ====================================== Extended abstracts for lightning talks and posters ====================================== * Extended abstracts of up to 3 pages (main text) * References, appendices, and acknowledgements are not included in this limit * Springer LNCS style (template available on the workshop website) ====================================== Important dates ====================================== * Extended Abstract Submission Deadline: August 12, 2026 * Notification of Acceptance: August 19, 2026 * Workshop Date: October 19, 2026 (afternoon) We welcome contributions from researchers working across the biological and artificial sciences and look forward to receiving your submissions. Best regards, Naoto Yoshida (Kyoto University) On behalf of the organizing committee: Ismael Freire ? Sorbonne Universit? Boris Gutkin ? ?cole Normale Sup?rieure Henning Sprekeler ? Technical University of Berlin Louis L?Haridon ? ENSEA / CNRS Lola Ca?amero ? CY Cergy Paris University Mehdi Khamassi ? Sorbonne Universit? --- Naoto Yoshida, PhD Program-Specific Researcher (Symbol Emergence Systems for Qualia Structures) Symbol Emergence Systems Lab. Graduate School of Informatics Research Bldg. 7 Kyoto University Yoshida-Honmachi, Sakyou-Ku, Kyoto, Japan. yoshida.naoto.8x at kyoto-u.ac.jp X: @movingsloth webpage: https://sites.google.com/view/movingsloth -------------- next part -------------- An HTML attachment was scrubbed... URL: From info at nisys.ai Thu Aug 6 08:20:42 2026 From: info at nisys.ai (Natural Intelligence) Date: Thu, 6 Aug 2026 14:20:42 +0200 Subject: Connectionists: Open Research Positions at NeuroAI start-up Natural Intelligence Message-ID: Dear all, *Natural Intelligence (NISYS GmbH)* is a deep-tech startup at the intersection of neuroscience and machine learning, founded by *Prof. Wolf Singer* and *Dr. Felix Effenberger*. We translate cutting-edge neuroscientific insight into robust, efficient next-gen AI: oscillatory recurrent architectures, self-organization, and learning rules that go beyond backpropagation, implemented in novel analog hardware platforms. We've closed a multi-million euro seed round, securing our funding for the coming years, and are currently building out our core R&D team. *Open roles:* full-time, on-site (Frankfurt), initially 3 years, several openings each: ? *Research Lead* - own a research direction and lead a group of scientists. ? *Senior Research Scientist* - design and test novel recurrent architectures from biological principles. The call for applications will remain open until early September 2026. We expect to review applications after the call closes and aim to get back to candidates in mid to late September. Full job descriptions at: www.nisys.ai Send applications and questions to hello at nisys.ai. Please pass this on to anyone who you think could be a good fit. -- Dr. Andr? Ferreira Castro Head of Research Operations Natural Intelligence (NISYS GmbH) -------------- next part -------------- An HTML attachment was scrubbed... URL: From announce at ucy.ac.cy Fri Aug 7 07:35:51 2026 From: announce at ucy.ac.cy (Announce) Date: Fri, 7 Aug 2026 11:35:51 +0000 Subject: Connectionists: International Conference on Software and Systems Reuse, Product Lines, and Configuration (VARIABILITY 2026): Call for Participation Message-ID: *** Call for Participation *** International Conference on Software and Systems Reuse, Product Lines, and Configuration (VARIABILITY 2026) 29 September - 2 October 2026, 5* St. Raphael Resort and Marina, Limassol, Cyprus https://conf.researchr.org/home/variability-2026 *** Early Registration Deadline: 31 August 2026 *** We are pleased to invite researchers, practitioners, educators, and students to participate in the inaugural edition of VARIABILITY 2026, the new flagship conference dedicated to variability engineering, software reuse, and software product line engineering. As modern software systems must cope with evolving requirements, heterogeneous platforms, changing user needs, adaptive behaviour, and increasingly AI-enabled development processes, variability engineering has become more important than ever. The conference provides a unique opportunity to discuss the latest advances in variability-enabled systems including modelling, analysis, automation, architectures, tools, empirical studies, software reuse, and software product line engineering. VARIABILITY 2026 builds upon the legacies of three leading conferences in the field: ? International Conference on Software Reuse (ICSR) ? International Working Conference on Variability Modelling of Software-Intensive Systems (VaMoS) ? International Systems and Software Product Line Conference (SPLC) By bringing these communities together, VARIABILITY 2026 creates a unified international forum for advancing the engineering, management, and exploitation of variability in software-intensive systems. Keynote Speakers VARIABILITY 2026 will feature keynote presentations by: ? Prof. Sigrid Eldh (Ericsson / M?lardalen University) ? Prof. Jean-Marc J?z?quel (University of Rennes, CNRS, Inria, IUF) Tracks VARIABILITY 2026 will feature presentations spanning many tracks including: ? Research ? Industry ? Journal-First ? Demonstrations and Tools ? Projects Showcase ? Doctoral Symposium Workshops ? MODEVAR - International Workshop on Languages for Modelling Variability https://modevar.github.io/ ? GAIV - Workshop on Generative AI and Variability https://sites.google.com/view/gaiv-2026 Tutorials ? Product Comparison using Feature Similarity Matching by Mike Mannion and Hermann Kaindl ? The Engineering and Manufacture of Software-based Products by Grady Campbell Awards The conference will honour excellence through: ? Awards for the most influential papers published more than 10 years ago in the predecessor conferences ? Best Research Paper and Best Industry Paper awards (sponsored by Springer) Join us in Cyprus for this historic inaugural edition and help build the future of variability- enabled systems and software! Organisation General Chairs ? George A. Papadopoulos, University of Cyprus, Cyprus ? Gilles Perrouin, FNRS & University of Namur, Belgium Research Track Chairs ? Thorsten Berger, Ruhr University Bochum, Germany ? Ina Schaefer, KIT, Germany Industry Track Chairs ? Shaukat Ali, Simula Research Lab and Oslo Metropolitan University, Norway ? Martin Becker, Fraunhofer IESE, Germany Journal First Track Chairs ? Mathieu Acher, University Rennes, Inria, CNRS, IRISA, France ? Xhevahire T?rnava, LTCI, T?l?com Paris, Institut Polytechnique de Paris, France Doctoral Symposium Track Chairs ? Rick Rabiser, LIT CPS, Johannes Kepler University Linz, Austria ? Iris Reinhartz-Berger, University of Haifa, Israel Demos and Tools Track Chairs ? Sandra Greiner, University of Southern Denmark, Denmark ? Leopoldo Teixeira, Federal University of Pernambuco Projects Showcase Chairs ? Daniel Struber, Chalmers, University of Gothenburg, Radbound University, Sweden ? Dalila Tamzalit, Nantes Universit?, France Hall of Fame Chairs ? Martin Becker, Fraunhofer IESE, Germany ? Goetz Botterweck, Lero - The Irish Software Research Centre and University of Limerick, Ireland ? Natsuko Noda, Shibaura Institute of Technology, Japan Workshops Chairs ? Lidia Fuentes, Universidad de Malaga, Spain ? Malte Lochau, University of Siegen, Germany Tutorials Chairs ? Loek Cleophas, Eindhoven University of Technology and Stellenbosch University, The Netherlands ? Mahsa Varshosaz, IT University of Copenhagen, Denmark Proceedings Chair ? Sophie Fortz, King's College London, UK Publicity Chairs ? Wesley Assun??o, North Carolina State University, USA ? Kentaro Yoshimura, Hitachi Ltd, Japan Local Organiser and Finance Chair ? George A. Papadopoulos, University of Cyprus, Cyprus -------------- next part -------------- An HTML attachment was scrubbed... URL: From giovanni.stanco at unina.it Fri Aug 7 11:31:46 2026 From: giovanni.stanco at unina.it (GIOVANNI STANCO) Date: Fri, 7 Aug 2026 15:31:46 +0000 Subject: Connectionists: [CFP: IWNC'26 - CNSM'26] Third International Workshop on Integrated Wireless Networking and Computing 2026 In-Reply-To: References: Message-ID: Third International Workshop on Integrated Wireless Networking and Computing 2026 (IWNC'26) Joint with the 22nd International Conference on Network and Service Management (CNSM 2026) Alcal? de Henares (Madrid), Spain // 26 - 30 October, 2026 https://sites.google.com/view/iwnc-2026/ CALL FOR PAPERS The Third International Workshop on Integrated Wireless Networking and Computing 2026 (IWNC'26) invites high-quality submissions of papers describing original and unpublished research results regarding the use of intelligent computing capabilities at the edge or within the network. TOPICS OF INTEREST Topics of interest include, but are not limited to: ? AI/ML-driven optimization in wireless systems ? Management and orchestration architectures and techniques for next-gen networks ? Orchestration solutions for control plane, core, and RAN ? Joint design of communication and computation functions ? Programmable wireless networks and smart NICs ? In-Network Computing architectures and applications ? Cross-layer and cross-domain design approaches ? Early-stage research ideas and system prototypes on Integrated Wireless Networking and Computing ? Testbeds, emulators, and benchmarking for wireless networks ? Wireless networks for AI applications ? AI and ML in Edge Computing ? Advancements in wireless networks for Federated Learning ? Resilience, scalability, and adaptability of next-gen networks ? Cybersecurity Challenges in Integrated Wireless Networking and Computing ? Security and Privacy in Wireless Networks ? Energy-aware and Sustainable Wireless Networks ? O-RAN and MEC solutions ? O-RAN orchestration and management SUBMISSION GUIDELINES Submitted manuscripts should use IEEE 2-column conference style and are limited to 6 pages (including references). All papers accepted by a workshop will be published in the CNSM 2026 Proceedings and will be sent for inclusion in the IEEE eXplore digital library. The submission and revision process will be managed through the EDAS system (submission link: https://edas.info/N35574). IMPORTANT DATES ? Workshop papers submission deadline: August 14, 2026 ? Notification of acceptance: September 11, 2026 ? Camera ready deadline: September 18, 2026 ? Workshop: October 30, 2026 We look forward to your valuable contributions and hope to see you at the workshop. Best regards, Stefania Zinno and Giovanni Stanco Third International Workshop on Integrated Wireless Networking and Computing 2026 (IWNC'26) -------------- next part -------------- An HTML attachment was scrubbed... URL: From moritz.grosse-wentrup at univie.ac.at Fri Aug 7 04:27:45 2026 From: moritz.grosse-wentrup at univie.ac.at (Moritz Grosse-Wentrup) Date: Fri, 7 Aug 2026 10:27:45 +0200 Subject: Connectionists: Postdoc Openings at the University of Vienna in Cognitive Computational Neuroscience and Brain-AI Interfacing Message-ID: <9c5c6e6d-fbcb-40c8-9256-cdd652b12607@univie.ac.at> Dear all, We have several postdoc openings in the Research Group Neuroinformatics at the University of Vienna. The positions will contribute to our research in cognitive computational neuroscience [1] and brain-AI interfacing [2], including the development of AI methods for offline and online neural decoding. We welcome applicants with a strong background in machine learning, computational neuroscience, neural signal processing, brain-computer interfaces, or related fields, as demonstrated through first-author publications in relevant high-quality journals (e.g., PLoS Comp. Bio, JNE, IEEE TBME / TNSRE, etc.) or at top conferences (e.g., NeurIPS, ICML, ICLR, etc.). The initial appointment is for up to two years, with the possibility of extension subject to funding and mutual agreement. Salary follows the Austrian Universities Collective Bargaining Agreement and depends on recognized prior experience; the expected gross annual salary is approximately ?70,000. Please send your application in a single PDF (including a cover letter, CV incl. a list of publications, and contact information for three references) to fsg.neuroinformatics at univie.ac.at. We will start screening applications on September 7. The positions will remain open until filled. Starting dates are flexible and can be negotiated. The University of Vienna has an anti-discriminatory employment policy and attaches great importance to equal opportunities, the advancement of women and diversity. We place special emphasis on increasing the number of women in senior and in academic positions. Given equal qualifications, preference will be given to female candidates. For any informal inquiries, please send me an email at moritz.grosse-wentrup at univie.ac.at. Best, Moritz 1. Grosse-Wentrup, Moritz, et al. "Neuro-cognitive multilevel causal modeling: A framework that bridges the explanatory gap between neuronal activity and cognition." PLoS Computational Biology 20.12 (2024): e1012674. 2. Meunier, Anja, et al. "A conversational brain-artificial intelligence interface." IEEE Transactions on Neural Systems and Rehabilitation Engineering (2026). -- Univ.-Prof. Dr.-Ing. Moritz Grosse-Wentrup Research Group Neuroinformatics Faculty of Computer Science University of Vienna Kolingasse 14-16, A-1090 Wien, Austria moritz.grosse-wentrup at univie.ac.at +43-1-4277-79610 -------------- next part -------------- A non-text attachment was scrubbed... Name: OpenPGP_signature.asc Type: application/pgp-signature Size: 665 bytes Desc: OpenPGP digital signature URL: From sabu.thampi at iiitmk.ac.in Fri Aug 7 12:46:53 2026 From: sabu.thampi at iiitmk.ac.in (Sabu M. Thampi) Date: Fri, 7 Aug 2026 22:16:53 +0530 Subject: Connectionists: =?utf-8?q?Call_for_Papers=3A_ICogSys_2026_?= =?utf-8?q?=E2=80=93_International_Conference_on_Cognitive_and_Inte?= =?utf-8?q?lligent_Systems?= Message-ID: *Call for Papers: ICogSys 2026 ? International Conference on Cognitive and Intelligent Systems* Where Cognitive AI Meets Intelligent Systems ? November 26?28, 2026 | ? Trivandrum, Kerala, India (Hybrid) ? Conference Website: https://qwi-events.org/icogsys2026/ The International Conference on Cognitive and Intelligent Systems (ICogSys 2026) invites original and unpublished research contributions presenting significant advances in Cognitive Artificial Intelligence, Machine Learning, and Intelligent Systems. ICogSys 2026 provides an international forum for researchers, academicians, industry professionals, practitioners, and research scholars to present innovative research, exchange ideas, and discuss emerging trends in intelligent computing and AI-enabled technologies. The conference aims to foster collaboration between academia, industry, and government while addressing current and future challenges in artificial intelligence and intelligent systems. *Topics of Interest* Authors are invited to submit original research contributions in, but not limited to, the following areas: Cognitive Artificial Intelligence Machine Learning and Intelligent Algorithms Computer Vision, Image and Signal Processing Trustworthy and Responsible AI Distributed Systems and Intelligent Infrastructure Cyber-Physical Systems, Robotics and Digital Twins AI for Public Infrastructure and Smart Cities Digital Governance and Public AI Systems AI Applications in Finance, Industry, Agriculture and Mobility Edge AI, Embedded Intelligence and Intelligent Hardware AI-Driven Cybersecurity and Cyber-Resilience Generative AI and Foundation Models AI for Healthcare and Biomedical Systems Green AI and Sustainable Computing *Proceedings Publication* The peer-reviewed and selected papers presented at ICogSys 2026 will be published by Springer in the prestigious Lecture Notes in Networks and Systems (LNNS) book series. The Springer Editorial Board has approved one proceedings volume for ICogSys 2026 in the LNNS series, comprising a maximum of 50 selected papers. The Lecture Notes in Networks and Systems (LNNS) series is indexed by Scopus, EI Compendex, INSPEC, WTI Frankfurt eG, zbMATH, and SCImago. In addition, all books published in the series are submitted for consideration for indexing in the Web of Science. All accepted papers will be subject to Springer's standard quality and integrity checks. Only papers that successfully satisfy these checks and comply with Springer's publication policies will be included in the final proceedings volume. *Review Process* All submissions will undergo a rigorous double-blind peer-review process coordinated by the Technical Program Committee. Each paper will be evaluated by at least two independent reviewers based on originality, technical quality, significance, relevance, and clarity of presentation. *Submission Guidelines* Papers must present original and unpublished research. Papers must not be under review by any other conference or journal. At least one author of each accepted paper must register and present the paper at the conference. Only accepted and presented papers that successfully complete Springer's quality and integrity checks will be included in the proceedings. *Important Dates* Paper Submission Deadline: August 20, 2026 Notification of Acceptance: October 1, 2026 Camera-Ready Submission & Author Registration: October 31, 2026 Conference Dates: November 26?28, 2026 *Recognition: *The Top 10 highest-rated papers, based on the peer-review process, will receive a special concession in the conference registration fee. *Paper Submission: *Submit your manuscript through the EDAS Conference Management System: https://edas.info/N35209 *Contact* General Enquiries: info.icogsys at gmail.com Submission Queries: chairs.icogsys at gmail.com We warmly invite researchers, academicians, practitioners, industry professionals, and research scholars from around the world to contribute to ICogSys 2026 and join us in Trivandrum, Kerala, India, for three days of high-quality technical discussions, knowledge exchange, and collaboration in the rapidly evolving fields of Cognitive Artificial Intelligence and Intelligent Systems. -------------- next part -------------- An HTML attachment was scrubbed... URL: From emergingtechnetwork.publicity at gmail.com Sat Aug 8 09:22:01 2026 From: emergingtechnetwork.publicity at gmail.com (=?UTF-8?Q?Gizem_G=C3=BCltekin_Varkonyi?=) Date: Sat, 8 Aug 2026 16:22:01 +0300 Subject: Connectionists: CCSS (Co-Sponsored by IEEE): The International Conference on Cybersecurity Systems, 27-30 October 2026 | Paris, France Message-ID: [Apologies if you got multiple copies of this invitation] 2026 International Conference on Cybersecurity Systems (CCSS 2026) https://css-conference.org/2026/ 27-30 October 2026 | Paris, France Hybrid Event and Technically Co-Sponsored by IEEE France Section *CCSS 2026 CFP:* The International Conference on Cybersecurity Systems (CCSS 2026) is a lively and inclusive gathering where researchers, academics, and industry professionals from across the globe come together to explore the latest advancements in Cybersecurity Systems and intelligent systems. Cybersecurity Systems received a lot of attention as AI continues to reshape industries and push the boundaries of what?s possible. CSS 2026 creates a welcoming space for innovation, collaboration, and the exchange of ideas in the Cybersecurity domain. CCSS 2026 serves as a premier platform for researchers, academics, and industry professionals to converge and explore the latest advancements, challenges, and applications in the dynamic fields of Cybersecurity Models and Systems. The conference provides a collaborative environment for the exchange of ideas, fostering innovation and promoting interdisciplinary research at the intersection of these transformative domains. We invite the submission of original papers on all topics related to Intelligent Systems for Cybersecurity, with special interest in but not limited to: ? Network and Internet Systems Security ? Information and Data Security ? Cybersecurity Applications ? Cryptography and Encryption ? Identity & Access Management (IAM) ? Cloud and Edge Security ? Cybersecurity in Healthcare ? Cybersecurity in Internet of Things (IoT) systems ? Digital Forensics Systems ? Federated Learning Systems ? Risk Management & Governance ? Edge and Distributed AI Security ? Cybersecurity Ethics and Governance ? AI and Cybersecurity ? Cybersecurity and Smart Cities ? Cybersecurity and Quantum Computing ? Threat Detection & Incident Response ? Cybersecurity in Emerging Technologies *Submissions Guidelines and Proceedings* Manuscripts should be prepared in 10-point font using the IEEE 8.5" x 11" two-column format. All papers should be in PDF format, and submitted electronically at Paper Submission Link. A full paper can be up to 8 pages (including all figures, tables and references). Submitted papers must present original unpublished research that is not currently under review for any other conference or journal. Papers not following these guidelines may be rejected without review. Also submissions received after the due date, exceeding length limit, or not appropriately structured may also not be considered. Authors may contact the Program Chair for further information or clarification. All submissions are peer-reviewed by at least three reviewers. Accepted papers will appear in the CCSS Proceeding, and be published by the IEEE Computer Society Conference Publishing Services and be submitted to IEEE Xplore for inclusion. *Important Dates:* - Paper Submission: *August 20, 2026 (Firm and Final)* - Notification to Authors: August 30, 2026 - Camera Ready Submission: September 15, 2026 *Contact:* Please send any inquiry on CCSS to : emergingtechnetwork at gmail.com -------------- next part -------------- An HTML attachment was scrubbed... URL: From david at irdta.eu Sat Aug 8 04:39:47 2026 From: david at irdta.eu (David Silva - IRDTA) Date: Sat, 8 Aug 2026 10:39:47 +0200 (CEST) Subject: Connectionists: GeMAIHc 2027: early registration August 27 Message-ID: <2014632741.257832.1786178387351@webmail.strato.com> ******************************************************** 1st INTERNATIONAL SCHOOL ON GENERATIVE & MULTIMODAL AI FOR HEALTHCARE GeMAIHc 2027 Milan, Italy February 1-5, 2027 https://gemaihc.irdta.eu/2027/ ******************************************************** Co-organized by: Human Technopole IRDTA ? Institute for Research Development, Training and Advice ****************************************************** Early registration: August 27, 2026 ****************************************************** SCOPE AI is rapidly redefining biomedical research, clinical decision-making, and healthcare delivery worldwide. GeMAIHc 2027 will provide participants with a comprehensive overview of state-of-the-art AI methodologies. Topics will include generative modelling, multimodal data integration and augmentation, large language models and clinical NLP, medical image synthesis and analysis, as well as evaluation, robustness, validation, ethical considerations, and regulatory aspects of AI systems in healthcare. GeMAIHc 2027 aims to become a leading international forum at the intersection of generative AI, multimodal learning, and healthcare innovation. The program will emphasize both methodological foundations and real-world clinical applications, combining theoretical insights with practical perspectives. This interdisciplinary event will feature 18 monographic three-hour courses, 2 keynote lectures, 3 scientific sessions, 1 symposium, and a hackathon. Leading academics and industry pioneers will share their expertise and perspectives with attendees. In-person interaction and networking will be central components of the event, while full remote participation will also be possible. ADDRESSED TO The program is open to PhD students and postdoctoral researchers in AI, data science, and biomedical disciplines, clinicians and healthcare professionals, industry practitioners and innovators, and policymakers and stakeholders in digital health. There are no formal academic prerequisites for participation. Researchers and professionals at all career stages are welcome. An international audience is expected, with backgrounds spanning computer science, medicine, engineering, mathematics, statistics, physics, and the social sciences. GeMAIHc 2027 will offer a unique opportunity to gain in-depth knowledge of a rapidly evolving field, to interact with experts across disciplines, to establish collaborations and expand professional networks, and to explore how AI is reshaping the future of healthcare. VENUE GeMAIHc 2027 will take place in Milan, one of Europe?s leading international centers for science, industry, fashion, and finance. The venue will be: Human Technopole Viale Rita Levi-Montalcini, 1 20157 Milan, Italy https://humantechnopole.it/en/ STRUCTURE Three parallel courses will run throughout the event. Participants will be free to choose the sessions they wish to attend and may move between courses at any time. A symposium will offer participants and companies the opportunity to present ongoing research or industrial developments through 10-minute oral presentations. The school will include a hackathon, during which participants will work in teams to tackle challenges in generative and multimodal AI for healthcare. All lectures will be video recorded and made available to participants for 45 days after the event. Full live online participation will be possible. Nevertheless, the organizers emphasize the importance of in-person interaction and networking in research training events of this kind. KEYNOTE SPEAKERS Deborah Estrin (Cornell University), Transforming Longitudinal Care with Digital Biomarkers and Therapeutics Lyle Palmer (Adelaide University), Deep Learning in Medicine: Some Lessons from Epidemiology and Genomics PROFESSORS AND COURSES (list to be completed) Pierre Baldi (University of California Irvine), [introductory/advanced] The AI-driven Healthcare of the Future David Buckeridge (McGill University), [intermediate] Multi-Modal Data Analysis in Population and Public Health Alejandro Frangi (University of Manchester), [intermediate/advanced] Virtual Patient Populations for In Silico Trials Using Generative AI Based on Multimodal Real-World Data Charles Friedman (University of Michigan), [introductory] Synergizing AI and Learning Health Systems to Transform Health Judy Wawira Gichoya (Emory University), [introductory/advanced] Combination in a Hands on Lab for Harmonizing Radiology Datasets Maryellen Giger (University of Chicago), [introductory/intermediate] Role of Data & Algorithms in Trustworthy Medical Imaging AI Casey Greene (University of Colorado), [introductory/advanced] Translating Generative AI into Healthcare from Pharmacogenomics to Foundation Models Tina Hernandez-Boussard (Stanford University), [intermediate/advanced] The AI Lifecycle in Healthcare: Evaluation, Deployment, and Responsible Implementation Jianying Hu (IBM Thomas J. Watson Research Center), [intermediate/advanced] Advanced Computing for Biomedical Research and Discovery Jayashree Kalpathy-Cramer (University of Colorado), [introductory] Toward Digital Twins: Multimodal AI for Imaging and Precision Medicine Alex John London (Carnegie Mellon University), [intermediate/advanced] Ethical and Scientific Challenges to Unlocking the Clinical Value of Artificial Intelligence Tom Pollard (Massachusetts Institute of Technology), tba Hoifung Poon (Microsoft Research), [advanced] Toward Virtual Patient: AI for Accelerating Medical Discovery Jian Tang (Mila-Qu?bec AI Institute), [introductory/advanced] Generative AI for Protein Design Peter van Ooijen (University of Groningen), [intermediate] Multimodal AI for Adaptive Radiotherapy: Tumor Segmentation, Uncertainty, and Explainability Karin Verspoor (Royal Melbourne Institute of Technology), [introductory/intermediate] AI Scientists and beyond: Roles for GenAI in Biomedical Research and Discovery Jelmer M. Wolterink (University of Twente), [intermediate/advanced] Data Representations in Health Digital Twinning: From Features to Foundation Models HT RESEARCH PERSPECTIVES Human Technopole will organize 3 scientific sessions: Probabilistic Models for (Medical) Image Analysis, by Jan Funke, Florian Jug, Federico Carrara, and Benjamin Salmon AI Models for Computational Biology, by Andrea Sottoriva, Michele Calabr?, and Manuel Dileo AI in Health Data Science, by Francesca Ieva, Andrea Ganna, Michela Carlotta Massi, Nicole Fontana, and Alessia Mapelli SYMPOSIUM A symposium will feature voluntary 10-minute oral presentations on ongoing research projects and industrial developments. Participants interested in presenting should submit a one-page abstract including title, authors, and summary to david at irdta.eu by January 8, 2027. HACKATHON Hands-on team activities will be organized around challenges in generative and multimodal AI for healthcare. The challenges will be released two weeks before the beginning of the school. A jury will evaluate the submissions, and the winners will be announced at the end of February 2027. Winning teams will receive a modest monetary prize, while runners-up will receive certificates of recognition. ORGANIZING COMMITTEE Jan Funke (Milan) Ilaria Guerini (Milan) Francesca Ieva (Milan) Carlos Mart?n-Vide (Tarragona, program chair) Michela Carlotta Massi (Milan, local chair) Santiago Montes (Tarragona, webpage) Sara Morales (Luxembourg, finances) David Silva (London, organization chair) REGISTRATION Registration is available at: https://gemaihc.irdta.eu/2027/registration/ The selection of six courses requested during registration is tentative and non-binding. This information will help estimate demand for logistical planning purposes. As venue capacity is limited, registrations will be processed on a first-come, first-served basis. Registration will close once capacity has been reached. Early registration is strongly recommended. FEES Registration fees include access to all school activities and lunches. Several early registration deadlines are available, and fees vary depending on the registration period. Fees are identical for on-site and online participation. ACCOMMODATION Accommodation suggestions will be provided in due time at: https://gemaihc.irdta.eu/2027/accommodation/ CERTIFICATE Participants will receive a certificate indicating 40 hours of academic activities. This certificate should be suitable for participants seeking ECTS recognition from their home institutions. SPONSORS Companies, institutions, and organizations interested in sponsoring the event may download the sponsorship leaflet from: https://gemaihc.irdta.eu/2027/sponsors/ QUESTIONS AND FURTHER INFORMATION david at irdta.eu ACKNOWLEDGMENTS Human Technopole Universitat Rovira i Virgili IRDTA ? Institute for Research Development, Training and Advice -------------- next part -------------- An HTML attachment was scrubbed... URL: From nanda.harishankar-krishna at mila.quebec Fri Aug 7 17:53:30 2026 From: nanda.harishankar-krishna at mila.quebec (Nanda Harishankar Krishna) Date: Fri, 7 Aug 2026 17:53:30 -0400 Subject: Connectionists: [CfP] NeurIPS 2026 Workshop on Foundation Models for the Brain and Body Message-ID: We are pleased to announce the second edition of the ?Foundation Models for the Brain and Body ? workshop at NeurIPS 2026 in Sydney, Australia. The workshop brings together researchers at the intersection of biosignals, neuroscience, machine learning, and embodied intelligence. We invite submissions of short papers on novel research in neuroscience, biosignal analysis, machine learning, and embodied AI, with a focus on foundation models and representation learning for neural, physiological, and behavioral data. We also invite interactive demo proposals showcasing novel methods, devices, systems, or applications. We broadly define biosignals as any temporal signal directly or indirectly generated by the brain or body. This includes neural recordings such as EEG, MEG, and multi-electrode arrays; physiological signals such as heart rate, respiration, and EMG; and behavioral signals such as pose, gaze, and movement. Submissions may focus on methods, applications, or theoretical insights that advance our understanding and modeling of neural, physiological, and behavioral signals in humans and animals. We particularly encourage work on scalable foundation models that generalize across tasks, subjects, modalities, or domains. Relevant topics include, but are not limited to: ? Large-scale pretraining on neural or physiological signals ? Scaling laws for biosignal foundation models ? Brain-computer interfaces and neural decoding ? Transfer learning across recording modalities ? Cross-subject and cross-task generalization ? Few-shot, zero-shot, and continual learning ? Multimodal integration of biosignals ? Wearable technology and real-world deployment ? Modeling behavioral signals such as pose, gaze, and movement ? Foundation models for movement, motor control, and EMG ? Embodied intelligence and sensorimotor representation learning ? Closed-loop and interactive systems, including BCI control and assistive robotics ? Scaling techniques from robotics and control applied to biosignal modeling ? Applications to health and clinical diagnosis ? Interpretability of biosignal representations ? Synthetic data generation and augmentation ? Human-AI collaboration for signal interpretation and labeling ? Open datasets, benchmarks, and model evaluation ? Evaluation protocols and standardized benchmarks across modalities ? Reproducibility and standardized reporting *Important Dates (AoE)* ? Paper Submission Deadline: *September 4, 2026* ? Demo Submission Deadline: *September 19, 2026* ? Paper Author Notification: September 29, 2026 ? Workshop Date: December 11 or 12, 2026, in Sydney, Australia Stay tuned for announcements about our invited speakers, panelists, and workshop schedule. For submission details and more information, please visit: https://brainbodyfm-workshop.github.io/ -------------- next part -------------- An HTML attachment was scrubbed... URL: From yj.choi at asu.edu Sun Aug 9 18:57:44 2026 From: yj.choi at asu.edu (YooJung Choi) Date: Sun, 9 Aug 2026 15:57:44 -0700 Subject: Connectionists: Postdoctoral Research Position in Automated Reasoning and Machine Learning Message-ID: *Postdoctoral Research Position in Automated Reasoning and Machine Learning* *School of Computing and Augmented IntelligenceArizona State University* The School of Computing and Augmented Intelligence (SCAI) at Arizona State University invites applications for a *Postdoctoral Research Scholar* position in Artificial Intelligence. The successful candidate will join a multi-institutional research project investigating new methods for integrating machine learning and logic-based reasoning to build trustworthy, explainable, and computationally efficient AI systems. The project spans several areas of AI, including knowledge representation and reasoning, automated deduction, probabilistic inference, uncertainty quantification, machine learning, and explainable AI. The successful candidate will have the opportunity to pursue foundational research while collaborating with leading researchers from academia and industry. *Research Areas*Candidates with expertise in one or more of the following areas are encouraged to apply: - Proof Theory and Automated Deduction - Logic Programming - Knowledge Representation and Reasoning - Probabilistic Inference and Graphical Models - Probabilistic Programming - Neuro-Symbolic AI *Qualifications*Applicants should have: - A Ph.D. in Computer Science, Artificial Intelligence, Machine Learning, Statistics, Mathematics, or a closely related field. - A strong publication record in leading AI conferences or journals (e.g., AAAI, IJCAI, KR, UAI, NeurIPS, ICML, ICLR, AISTATS, CADE, LICS, ICAPS). - Demonstrated research expertise in one or more of: - Knowledge representation and automated reasoning - Proof theory, theorem proving, or logic programming - Probabilistic inference, probabilistic graphical models, or uncertainty in AI - Neuro-symbolic AI or mathematical logic - Strong programming skills in Python and experience developing research software. Experience with one or more of the following is desirable, but not required: - Bayesian methods or Gaussian Processes - Tractable probabilistic inference or probabilistic circuits - Automated theorem proving or symbolic reasoning systems - Explainable AI and uncertainty estimation *Research Environment* The position is based at Arizona State University and involves collaboration with researchers from academia and industry on foundational AI research with applications to trustworthy intelligent systems. The successful candidate will work closely with Prof. Siddharth Srivastava, Prof. YooJung Choi, and Prof. Giulia Pedrielli at ASU. Arizona State University has a vibrant AI research community spanning knowledge representation, planning, machine learning, robotics, formal methods, and human-AI collaboration. The position offers opportunities to publish in leading AI venues and contribute to the development of next-generation neuro-symbolic AI systems. *Appointment* - Full-time postdoctoral appointment - Initial one-year appointment, renewable based on performance and funding - Competitive salary and benefits - Preferred start date: Fall 2026 *Application*Applicants should submit: - Curriculum vitae - A one-page research statement describing research interests and relevant experience - Contact information for three references - Up to three representative publications Review of applications will begin immediately and continue until the position is filled. Questions about the position may be directed to: scale_ml_lp at googlegroups.com. -------------- next part -------------- An HTML attachment was scrubbed... URL: From taro.toyoizumi at riken.jp Sun Aug 9 23:35:30 2026 From: taro.toyoizumi at riken.jp (Taro Toyoizumi) Date: Mon, 10 Aug 2026 03:35:30 +0000 Subject: Connectionists: Neural Networks : Volume 203 In-Reply-To: <0100019fe7e25c62-2fed93c8-10d4-4112-b29e-2e298f88cac1-000000@email.amazonses.com> References: <0100019fe7e25c62-2fed93c8-10d4-4112-b29e-2e298f88cac1-000000@email.amazonses.com> Message-ID: [wordmark] New Issue: Neural Networks New issue available on ScienceDirect [Cover Image Neural Networks] Neural Networks Volume 203, November 2026 Editorial Board Article Number 109480 Reviews ________________________________ Diffusion models for hyperspectral image analysis: A comprehensive review Article Number 109109 Xing Hu, Xiangcheng Liu, Qianqian Duan, Lian Zhang, Huiliang Shang, Linhua Jiang, Haima Yang, Dawei Zhang Cognitive Science ________________________________ R&B - rhythm and brain: Cross-subject decoding of music from human brain activity Article Number 109195 Matteo Ciferri, Matteo Ferrante, Nicola Toschi Attentive pre-training question embeddings for knowledge tracing with semantically-enhanced knowledge structure and concept label-guided heterogeneous graph representation Article Number 109194 Jinjie Zhou, Senlin Luo, Songling Wu, Xiaonan Yang, Limin Pan, Deshan Yang Neuroscience ________________________________ Multi-?-stability and fixed-time multistability of switched fuzzy neural networks with discontinuous activation functions Article Number 109101 Zhenxue Lu, Shiqin Ou, Zhenyuan Guo, Xiaobing Nie, Shiping Wen Learning Systems ________________________________ Unsupervised fine-tuning of vision-language models by fusing classifier tuning and visual prompt tuning Article Number 109082 Wenyang Chen, Zhanxuan Hu, Yonghang Tai, Feiping Nie VPAF-FSCIL: Virtual prototype calibration and parameter-adaptive freezing for few-shot class-incremental learning Article Number 109094 Ye Yao, Junxi Li, Xiong Chen HIVE: A hypergraph-based game-theoretic interactive value decomposition engine for multi-lateral agents collaboration Article Number 109103 Changdong Zhou, Ruofan Hu, Naiyao Wang, Ke Song, Hongbo Liu Adaptive critic designs for event-based multi-agent systems with asymmetric constraints Article Number 109100 Wenting Yan, Ding Wang, Xinrui Ma, Junfei Qiao KAB: A knowledge-aligned benchmark for reproducible evaluation of distantly supervised relation extraction Article Number 109081 Bowen Liu, Junhang Hu, Yucong Lin, Hong Song, Yaqing Nie, Hongmin Xiao, Zichao Lin, Jingtao Li, Xutao Weng, Zhaoli Su, Jinfu Li, Jian Yang An end-to-end dual backbone framework for UAV fish disease detection using optimized GateHCO feature fusion mechanism Article Number 109140 Jiaji Pan, Zehui Liu, Xuanzhi Li, Xianwu Luo, Fengming Huang, Miaolei He, Wei Qin, Yaoyi Cai, Jun Xiao, Yu Li, Hao Feng MaskCtrl: Training mask networks as self-explainable and performant controllers via deep reinforcement learning Article Number 109107 Shi Peng, Si Liu, Dapeng Zhi, Peixin Wang, Min Zhang Negative prompt-guided optimization: Enhancing soft prompt generalization in vision-language models Article Number 109093 Suneung Kim, Seong-Whan Lee A bio-inspired neuromorphic system for fusing visual features and autonomous learning Article Number 109097 Mei Guo, Yaoyao Zi, Lixin Liu, Jikang Liu, Qiye Yang, Jingzhi Xu, Gang Dou, Lihua Wang, Da Chen Graphical abstract [http://ars.els-cdn.com/content/image//1-s2.0-S0893608026005575-ga1.jpg] Dynamic adaptive multi-view contrastive learning for unsupervised person re-identification Article Number 109083 Zhi-Hua Li, Xue-Yan Wang, Si-Bao Chen, Chris H.Q. Ding, Bin Luo Window-to-window BEV representation learning for limited FoV cross-view geo-localization Article Number 109099 Lei Cheng, Daikun Liu, Lingquan Meng, Teng Wang, Changyin Sun Cross-modal recipe retrieval via multi-granularity alignment Article Number 109102 Runqi Zan, Tao Yao, Yuxin Yu, Guorui Sheng, Zongchao Huang Subgraph-Mamba: Subgraph Mamba model with positional encoding Article Number 109113 Denggao Qin, Xianghong Tang, Jianguang Lu, Jing Yang, Philip S. Yu Cross attention-based prior deformation for category-level 6D pose estimation Article Number 109096 Shuai Guo, Yongchao Yang, Lifeng Zhang, Chunge Cao, Yazhou Hu Memory-efficient divide-and-conquer attention for lightweight image super-resolution Article Number 109085 Rui He, Zhenyang Zhu, Xiaoyang Mao HyperNATE: Scaling tensor-based hypergraph neural networks through attention Article Number 109139 Nicol?s Bello, Fuli Wang, Daniel L. Lau, Gonzalo R. Arce Graphical abstract [http://ars.els-cdn.com/content/image//1-s2.0-S0893608026006003-ga1.jpg] Scattering center guided mono-static radar cross section prediction Article Number 109061 Zehao Tang, Yuxuan Luo, Shuo Zhang, Biao Leng QMSANet: A quaternion multi-scale attention network for robust color image denoising Article Number 109091 Yi Liu, Qi Xie, Yu Guo, Guoqing Chen, Boying Wu, Deyu Meng, Jean-Michel Morel, Qiyu Jin, Michael Kwok-Po Ng Memristor-based reconfigurable architecture for binarized neural networks: Implementation and robustness analysis Article Number 109104 Xiaoyang Liu, Xu Xie, Banghu Yin, Rusheng Ju Mapp: A model-agnostic privacy-preserving framework for two-party graph neural network inference Article Number 109133 Juxiang Zeng, Pinghui Wang, Yangchao Qian, Tingqin Liu, Jing Tao, Xiaohong Guan Robust zero-shot learning with ambiguous labels via visual-semantic alignment and dynamic disambiguation Article Number 109142 Jiangnan Li, Xiaowen Yan, Linqing Huang, Jinfu Fan SGNet: Spectral-geometric neural network for structured representation learning Article Number 109135 Idowu Paul Okuwobi, Jingyuan Liu, Olayinka Susan Raji, Olusola Funsho Abiodun Asymmetric drug-drug interaction prediction based on diffusion-augmented graph attention network Article Number 109134 Lei Zhang, Fan Yang, Jie Xia, Kaibiao Lin, ZhaoRi Guo Robust network pruning for enhanced accuracy under perturbations via structural and distributional consistency Article Number 109152 Wenhui Shi, Jiangang Yang, Yangbin Xu, Xiaoran Xu, Wenyue Chong, Jian Liu Focus on the essentials: Learning to attend to the most critical information for visual question answering Article Number 109154 Kun Zeng, Zhixin Li URSMamba: Universal remote sensing image steganography using state space model Article Number 109132 Chao Yang, Shiyuan Wang, Ying Huang, Mingqiang Guo Randomized neural network with adaptive forward regularization for online task-free class incremental learning Article Number 109115 Junda Wang, Minghui Hu, Ning Li, Abdulaziz Al-Ali, Ponnuthurai Nagaratnam Suganthan OACodec: Audio attribute disentanglement via orthogonal disentanglement and mutual information minimization Article Number 109128 Yukun Qian, Wenjie Zhang, Zehua Zhang, Lianyu Zhou, Xuyi Zhuang, Mingjiang Wang Cross-domain sequential recommendation via interest-guided knowledge migration Article Number 109158 Xueting Li, Shaoqing Wang, Yan Wang, Haoran Yang, Fuzhen Sun Hybrid Mamba-CNN network for forward-looking sonar image segmentation with acoustic background suppression mechanism Article Number 109165 Hu Xu, Ju He, Haoran Hu, Guoqing Xie, Yang Yu Projection with mixed-size anchor graphs Article Number 109149 Qianyao Qiang, Bin Zhang, Jason Chen Zhang, Chaodie Liu, Feiping Nie Pareto-optimal estimation and policy learning for balancing short-term and long-term outcomes Article Number 109144 Yingrong Wang, Anpeng Wu, Haoxuan Li, Weiming Liu, Baohong Li, Qiaowei Miao, Ruoxuan Xiong, Fei Wu, Kun Kuang EndoUFM: Utilizing foundation models for monocular depth estimation of endoscopic images Article Number 109141 Xinning Yao, Bo Liu, Bojian Li, Jingjing Wang, Jinghua Yue, Fugen Zhou MGCFI-Net: Multi-scale globally aware feature learning with cross-view feature interaction for multi-view stereo Article Number 109137 Ming Han, Hui Yin, Aixin Chong, Hua Huang Graphical abstract [http://ars.els-cdn.com/content/image//1-s2.0-S0893608026005988-ga1.jpg] ED-SAM: Sharpness-aware minimization with energy-adjusted perturbations and direction-corrected updates Article Number 109185 Hailiang Ye, Xinyi Fang, Ming Li, Feilong Cao FedEBM: Robust graph federated learning via energy-based model Article Number 109193 Jiayu Wang, Jinyan Wang, Zeming Gan, Xianxian Li, De Li GAFGCN: A graph augmented fusion network for enhanced deep clustering in attributed graphs Article Number 109173 Yingming Jiang, Haiyan Guo GRLT: Learning more from teachers by rethinking knowledge distillation from GNNs to MLPs Article Number 109170 Yaogang Geng, Hong Yu, Guoyin Wang, Yongfang Xie Enhance graph alignment for large language models Article Number 109162 Haitong Luo, Xuying Meng, Suhang Wang, Tianxiang Zhao, Fali Wang, Yujun Zhang Graph knowledge distillation with high frequency in homophily and heterophily graphs Article Number 109171 Donghai Guan, Junqiong Nie, Weiwei Yuan Graphical abstract [http://ars.els-cdn.com/content/image//1-s2.0-S0893608026006325-ga1.jpg] A unified multi-stream diffusion framework for robust video camouflaged object detection Article Number 109169 Yuyao Ke, Rui Yao, Kunyang Sun, Hancheng Zhu, Xixi Li, Jiaqi Zhao, Bing Liu CWITrack: Transformer tracking via local-global cross-window interaction Article Number 109181 Yuanyun Wang, Pengcheng Sha, Shenmiao Jin, Yichao Li, Jun Wang DPC: Dynamic purification chain for adaptive adversarial defense Article Number 109189 Zeshan Pang, Yuyuan Sun, Rongtao Liao, Xuehu Yan, Shasha Guo, Yuliang Lu Learning fair graph representation through graph information disentanglement Article Number 109184 Qingfeng Chen, Wujie Wei, Debo Cheng, Chuxun Liu, Jinyi Jie, Jiangzhang Gan, Shichao Zhang Ultralightweight progressive feature disentanglement and recomposition network for hyperspectral image classification Article Number 109200 Delong Kong, Shichao Zhang, Xiang Yu, Yanshuang Lu, Shanshan Yang, Jiahua Zhang Fine-grained hierarchical multi-round iterative semantic optimization attack method for RAG systems Article Number 109192 Qidong Chen, Vasile Palade, Zihao Yu, Ruixiang Deng, Jun Sun, Hao Wu Quadratic effects of linear interpolation between permutation-aligned neural networks Article Number 109074 Beatrix Benk? Auto-labeling for single-photon LiDAR semantic understanding under varying acquisition conditions Article Number 109188 Ziting Wen, Zili Zhang, Kemi Ding, Xiaoqiang Ren SAGE: Semantic-guided framework with decoupled optimization for open-vocabulary video visual relationship detection Article Number 109183 Shiqi Wang, Weiying Xue, Shuyi Hu, Haowen Li, Qi Liu EDA-OCBLS: An error-distribution aware one-class broad learning system for anomaly detection Article Number 109177 M. Tanveer, A. Mishra, A. Quadir, M. Sajid Self-training framework based on multi-granularity local cores for class-imbalanced semi-supervised classification in business applications Article Number 109145 Junnan Li, Xiaosheng Su, Leo Wang, Jicheng Ma, Yingjun Xia, Yuqing Gu, Shun Fu, Wenli Xu, Ziqiang He BigOrthoATD.Net: A scalable and adaptable distributed deep learning framework for multi-class orthopedic classification across imaging modalities in low-resourced settings Article Number 109190 Haider A. Alwzwazy, Laith Alzubaidi, Zehui Zhao, Ross Crawford, Omar Alnaseri, Raja Jurdak, Yuantong Gu HetDualCL: Dual-encoder contrastive learning for heterogeneous graphs Article Number 109205 Yangding Li, Jiawei Chai, Wenjie Zhang, Changwei Li, Xiangchao Zhao, Bingbing Xu, Shichao Zhang ESIMCE: Efficient and simple incomplete multi-view clustering via ensembles Article Number 109167 Haiyan Cheng, Hao Huang, Haiyan Wang, Jinrong Cui When task performance deceives: Task-geometry decoupling in learnable-curvature hyperbolic GNNs Article Number 109172 Lixian Chen, Jingchao Wang, Zhaorong Dai, Hanqian Liu, Danxiang Ai, Yang Shi Graphical abstract [http://ars.els-cdn.com/content/image//1-s2.0-S0893608026006337-ga1.jpg] Revisiting weakly supervised tabular anomaly detection from a cell-level perspective Article Number 109202 Jiahui Wang, Zhen Peng, Xujing Jia, Qika Lin, Lan Ma, Bin Shi Symplectic convolutional neural networks Article Number 109208 S?leyman Y?ld?z, Konrad Janik, Peter Benner Robust and efficient learning with granular ball support vector regression Article Number 109174 Reshma Rastogi, Ankush Bisht, Sanjay Kumar, Suresh Chandra Mathematical and Computational Analysis ________________________________ Unsupervised feature selection via anomaly-aware fuzzy graph fusion and diffusion multi-centroid learning Article Number 109108 Zhouqing Yan, Ziping Ma, Jinlin Ma, Huirong Li DGVAC: Disentangled Gaussian-vMF alignment for deep clustering Article Number 109114 Tangjun Ruan, Nanjun Yu, Shangshang Zhao, Feiyu Chen, Yi Huang ConMGIN: Interpretable multilayer GIN-Bayesian framework for spatial domain analysis Article Number 109155 Jie Li, Farong Liu, Haoyang Lv, Sheng Ren, Xin Gao, Bin Yu MoST: A monotone set transformer for scalable and verifiable neuro-fuzzy aggregation Article Number 109153 Jih-Jeng Huang, Chin-Yi Chen Identifying the seizure onset zone with phase-amplitude coupling Article Number 109151 Junfeng Lu, Denghai Wang, Dandan Kong, Kunying Meng, Rui Zhang, Hong Wan, Mingming Chen JointRel: Joint semantic embedding with relational message passing for knowledge graph completion Article Number 109156 Yunong Zhang, Jiashuang Huang, Weiping Ding, Min Xu CollDTI: Dual-encoder collaborative learning for drug-target interaction prediction Article Number 109175 Wanchen Li, Junlin Xu, Yajie Meng, Xunkun Cheng, Li Ye, Yinhui Jiang, Shuting Jin XAI-DTBD: Explainable dynamic threshold-based backdoor detection in graph neural networks Article Number 109157 Adil Ahmad, Anwar Shah, Muhamamd Adnan, Chau Yuen Fuzzy reinforcement learning synchronization of stochastic dynamic networks: An adaptive event-triggered strategy Article Number 109180 Jiayi Cai, Jianwen Feng, Jingyi Wang, Chengbo Yi, Guanrong Chen FNE-RTDETR: A lightweight end-to-end model for small-area tomato leaf disease detection and fine-grained classification Article Number 109179 Abudukelimu Abulizi, Junxiang Ye, Mayilamu Musideke, Gengrong Zhang, Youli Su, Yajun Zhang, Halidanmu Abudukelimu Engineering and Applications ________________________________ DDFL: dual defense against poisoning attacks in privacy-preserving federated learning Article Number 109095 Cheng Guo, Moyan Tian, Xueguang Li, Hui Sun, Yingmo Jie Pose-guided YOLO-CED: A two-stage framework for robust SMD-PCB defect detection in mobile manipulators Article Number 109136 Mingxiao Sun, Qiang Zhao, Qiuyu Zhang, Tiantian Luan Stochastic approximation to contrastive learning Article Number 109098 Erland Brandser Olsson, Zhirong Yang NSFF: Noise and semantic features fusion for AI-generated image detection Article Number 109130 Haoran Yang, Ruiqiang Ma, Gang Wang, Jien Kato Graphical abstract [http://ars.els-cdn.com/content/image//1-s2.0-S0893608026005915-ga1.jpg] Global-focal adaptation with information separation for noise-robust transfer fault diagnosis Article Number 109112 Junyu Ren, Wensheng Gan, Guangyu Zhang, Wei Zhong, Philip S. Yu MFF-M3AD: A unified reconstruction method with multi-scale feature fusion for multi-category 3D anomaly detection Article Number 109131 Hanzhe Liang, Chenxi Hu, Yejin Tang, Linlin Shen, Jinbao Wang, Can Gao Better utilization of illumination prior via KANs for nighttime flare removal Article Number 109138 Aoxiang Ning, Minglong Xue, Senming Zhong, Palaiahnakote Shivakumara, Mingliang Zhou Collective reflection-based multi-agent reinforcement learning framework for task-oriented dialogue policy learning Article Number 109110 Kai Xu, Zhenyu Wang, Yangyang Zhao, Bopeng Fang Heterogeneous neural blind deconvolution: A signal processing-empowered foundation feature extractor for bearing fault diagnosis Article Number 109143 Jing-Xiao Liao, Chao He, Jipu Li, Xiao-Cong Zhong, Jinwei Sun, Yiu-ming Cheung, Feng-Lei Fan, Shiping Zhang, Xiaoge Zhang Different direction adversarial sample for diffusion model Article Number 109196 Shan He, Hai Da, Jun Jiang, JiaYang Li, FuGui Chen TextEconomizer: Enhancing lossy text compression with denoising transformers and entropy coding Article Number 109111 Mahbub E Sobhani, Anika Tasnim Rodela, Chowdhury Mofizur Rahman, Dewan Md. Farid, Swakkhar Shatabda Lightweight spiking transformer towards neurodynamic integration framework Article Number 109127 Miao Miao, Haoyan Liu, Shurui Fan, Kewen Xia Graphical abstract [http://ars.els-cdn.com/content/image//1-s2.0-S0893608026005885-ga1.jpg] AADFNet: An adaptive asymmetric dual-branch fusion network for background-robust grasping Article Number 109150 Tao Fan, Chenyang Liu, Qiuyang Dai, Fang Fang AnyDesign: Versatile area fashion editing via mask-free diffusion Article Number 109129 Yunfang Niu, Dong Yi, Lingxiang Wu, Jie Peng, Jinqiao Wang Omni-domain energy-efficient decision-making for large-scale heterogeneous platoons with dual-level graph reinforcement learning Article Number 109148 Xin Gao, Changjian Zhao, Xueyuan Li, Ao Li, Zhaoyang Ma, Zirui Li Image restoration model compression via mamba-oriented heterogeneous knowledge distillation Article Number 109159 Sai Yang, Bin Hu, Xiaoxin Wu, Fan Liu, Wanzhi Wen Large language model augmented framework with domain-specific knowledge integration for medical named entity recognition Article Number 109178 Haochen Zou, Yongli Wang Solver-in-the-loop joint operator learning: Fractional Laplace-Beltrami features for interface reconstruction Article Number 109067 Yangyang Zheng, Huayi Wei, Shuhao Cao, Ruchi Guo In-vehicle low-light face image enhancement with physical-semantic constrained diffusion and gated selective state-space scanning Article Number 109161 Pancheng Zhang, Zhe Chen, Yihui Hu Graphical abstract [http://ars.els-cdn.com/content/image//1-s2.0-S0893608026006222-ga1.jpg] Debiased medication recommendation through fusing frequent pattern and temporal medical records Article Number 109168 Xiaobo Li, Xiaodi Hou, Simiao Wang, Shilong Wang, Xiaokun Zhang, Yijia Zhang Projective and functional projective synchronizations in quaternion-valued inertial neural networks with interaction term through novel generalized time-varying delay lemma and one-norm method: Application in secure communication Article Number 109164 Vaibhav Agrawal, Ahindra Sarkar, Subir Das, Vineet Kumar Singh Point-wise conditional diffusion models for physical systems with shape variations: Applications to spatio-temporal and large-scale systems Article Number 109186 Jiyong Kim, Sunwoong Yang, Namwoo Kang ESTGFormer: A spatio-temporal graph transformer with embedding and structure-aware loss for traffic forecasting Article Number 109182 Famiao Mou, Zhineng Lv, Xuesong Jin, Lijun Yun, Zaiqing Chen Diffusion model with Rician?Gaussian priors for robust MR image synthesis Article Number 109163 Hyoyoung Jang, Sangmin Lee Hypergraph mixer WaveNet: A lightweight spatio-temporal framework utilizing hypergraphs for high-order spatial modeling in traffic forecasting Article Number 109176 Nguyen-Huu An, Dung-Cam Quang, Van-Vang Le Efficient FPGA accelerator for low-power high-speed BCI motor imagery classification using novel deep learning Article Number 109105 Saravanakumar C, Srinivasan C, Immaculate Joy S SAS-bench: A fine-grained benchmark for evaluating short answer scoring with large language models Article Number 109199 Peichao Lai, Kexuan Zhang, Yi Lin, Linyihan Zhang, Feiyang Ye, Jinhao Yan, Yanwei Xu, Conghui He, Wentao Zhang, Yilei Wang, Bin Cui Read the full issue on ScienceDirect -------------- next part -------------- An HTML attachment was scrubbed... URL: From ahmed.imprs at gmail.com Mon Aug 10 04:21:53 2026 From: ahmed.imprs at gmail.com (Ahmed El Hady) Date: Mon, 10 Aug 2026 10:21:53 +0200 Subject: Connectionists: Postdoctoral Opportunity Zukunftskolleg Message-ID: <8507eb1b-d062-423e-bf5a-f649cf94839a@gmail.com> *_Four 2-year Postdoctoral Fellowships at the Zukunftskolleg_* (Full-time, E 13 TV-L) *Reference No: 2026/138 The preferred start date is April 1st, 2027. Conditionally on the submission of an external grant proposal, the position may be extended for an additional year. * The University of Konstanz is a dynamic and internationally successful research university with approximately 10,000 students. The campus university?s architecture promotes interdisciplinary cooperation and a sense of community among researchers, lecturers and students. The University has been continuously funded by the German Excellence Strategy since 2006. The /Zukunftskolleg/?is an Institute for Advanced Study for early career postdoctoral researchers in the natural sciences, humanities and social sciences. The /Zukunftskolleg offers postdoctoral researchers/?Independence for their research, interdisciplinary and intergenerational exchange as well as intra-university integration and international contacts. The working language is English. Each fellow is a member of both a university department and the Zukunftskolleg. Additionally, fellows of the Zukunftskolleg can be affiliated to the Cluster of Excellence at the University of Konstanz: *"Politics of Inequality "*?or?"*Centre for the Advanced Study of Collective Behaviour*" . The target group consists of excellent early career postdoctoral researchers who demonstrate scientific excellence, research independence, and a strong capacity for teamwork, indicating high potential for a successful academic career The applications are evaluated in a two-stage process by international reviewers. The candidates will be informed about the decision around mid-February 2027. /I am happy to discuss potential projects, for those interested to apply,? on the interface of computational sciences, theoretical physics and collective behavior and application of these approaches to the myriad of collective behavior data available in Konstanz. /Please drop an email to ahmed.el-hady at uni-konstanz.de All details can be found here:?link: https://stellen.uni-konstanz.de/jobposting/47ea117d40ac0aa2409ab5ecf4ed18ee0afb70cc All the best regards, Ahmed Dr. Ahmed El Hady Research Group leader Centre for Advanced Study of Collective Behavior University of Konstanz Max Planck Institute of Animal Behavior Universit?tsstra?e 10, 78464 Konstanz Room ZT 907 ,Postbox 687 -------------- next part -------------- An HTML attachment was scrubbed... URL: From annecollins at berkeley.edu Mon Aug 10 12:09:47 2026 From: annecollins at berkeley.edu (Anne Collins) Date: Mon, 10 Aug 2026 12:09:47 -0400 Subject: Connectionists: Faculty search - UC Berkeley psychology Message-ID: UC Berkeley psychology will be having a faculty search for an assistant teaching professor in quantitative methods, with prospect to equivalent of tenure. Add is linked below - please pass along to potential candidates. https://aprecruit.berkeley.edu/JPF05480 Anne Collins Associate Professor Department of Psychology University of California, Berkeley (510) 664-7146 -------------- next part -------------- An HTML attachment was scrubbed... URL: From genevievejourdainlambert at gmail.com Tue Aug 11 04:36:56 2026 From: genevievejourdainlambert at gmail.com (Genevieve Jourdain-Lambert) Date: Tue, 11 Aug 2026 10:36:56 +0200 Subject: Connectionists: =?utf-8?q?=5BCOMPLEX_NETWORKS_2026=5D_=5BCFP=5D_J?= =?utf-8?q?oin_us_in_Granada_=E2=80=94_submit_by_September_2?= Message-ID: Dear colleagues, We would like to remind you that submissions are open for COMPLEX NETWORKS 2026, the 15th International Conference on Complex Networks and Their Applications. ? Granada, Spain ? Conference: December 2?4, 2026 ? Tutorials: December 1, 2026 ? Submission deadline: September 2, 2026 The conference welcomes contributions on network science, complex systems, graph learning, computational social science, and applications of complex networks across disciplines. ? Submission formats ? Full papers, up to 12 pages ? Extended abstracts, up to 4 pages ? Publication ? Accepted full papers will be published in the Springer proceedings ? Accepted extended abstracts will be published in the Book of Abstracts with ISBN ? Selected contributions will be considered for journal special issues ? Keynotes Alexandra Brintrup, Nitesh Chawla, Kimmo Kaski, Bruno Lepri, Eckehard Sch?ll ? Tutorials Fariba Karimi, Huijuan Wang ? Website https://complexnetworks.org/ ? Submission https://complexnetworks.org/submission/ We warmly invite you to submit your work and to share this call with colleagues, students, postdocs, and young researchers who may be interested. Best regards, The COMPLEX NETWORKS 2026 Organizing Committee -------------- next part -------------- An HTML attachment was scrubbed... URL: From risto at cs.utexas.edu Mon Aug 10 19:42:17 2026 From: risto at cs.utexas.edu (Risto Miikkulainen) Date: Mon, 10 Aug 2026 16:42:17 -0700 Subject: Connectionists: Call for Contributions to the Agentic Web workshop at NeurIPS 2026 Message-ID: We are pleased to announce a new workshop at NeurIPS 2026 in Atlanta: The Agentic Web: Decentralized, Continually-Adapting Agent Ecosystems The workshop focuses on the question: What new science do we need for agentic intelligence at internet scale? The premise is that AI's basic unit of computation is shifting---from single models and individual agents toward continually running, internet-scale collectives of agents that discover one another, communicate, coordinate, and adapt over open networks. We invite submissions that address issues that arise, including: - How to represent and discover agent capabilities at web scale - How learning, credit assignment, and self-improvement can take place over continually changing agent networks - How large populations of agents can coordinate, remain trustworthy, and stay safe under strategic pressure. The workshop welcomes short papers (4 pages) and long papers (9 pages) in the NeurIPS format through OpenReview, with the deadline of September 5th AoE. For more details see https://projectnanda.org/workshops/neurips26 (and https://theagenticweb.ai in the near future). - Pradyumna Chari, Eric Horvitz, Zixuan Ke, Risto Miikkulainen -------------- next part -------------- An HTML attachment was scrubbed... URL: From aong at flatironinstitute.org Tue Aug 11 11:14:51 2026 From: aong at flatironinstitute.org (Allison Ong) Date: Tue, 11 Aug 2026 11:14:51 -0400 Subject: Connectionists: Start date extended! Joint junior faculty position in Computational Neuroscience Message-ID: The Graduate Center of the City University of New York (CUNY), and the Center for Computational Neuroscience at the Flatiron Institute, invite applications for a joint position, with start date of August 2026. We seek an innovative scholar whose research integrates computational approaches with empirical neuroscience to investigate the mechanisms underlying cognition. The ideal candidate will have a strong record of research in perception, planning, decision-making, memory, or animal communication and/or human language, and a doctoral degree in Physics, Math, Engineering, Neuroscience, Cognitive Science, Psychology, or Computer Science. The CUNY appointment will be in the M.S. Program in Cognitive Neuroscience. Responsibilities include research, teaching, mentoring, and service. The candidate is expected to generate an active, externally funded research program, and to participate in graduate teaching and research training. Teaching duties may include courses in computational methods and advanced courses in neuroscience. Candidates should demonstrate a commitment to interdisciplinary collaboration and to mentoring graduate students in a diverse and inclusive academic environment. The Flatiron appointment will be a Visiting Scholar position in the Center for Computational Neuroscience (CCN). **The start date has been extended to Jan 2027.** Applications must be submitted to both positions - these can be identical, although applicants may wish to highlight specific aspects of their experience or interests in the cover later for each submission. https://cuny .jobs/new-york-ny/assistant-professor-ms-in-cognitive-neuroscience/51674D0A749E42818AB89D06270999B5/job/ https://www.simonsfoundation.org/flatiron/careers/?tab=job-openings¢er=ccn -- *Eero Simoncelli* Scientific Director, Center for Computational Neuroscience Flatiron Institute, Simons Foundation -------------- next part -------------- An HTML attachment was scrubbed... URL: From emergingtechnetwork.publicity at gmail.com Tue Aug 11 11:23:11 2026 From: emergingtechnetwork.publicity at gmail.com (=?UTF-8?Q?Gizem_G=C3=BCltekin_Varkonyi?=) Date: Tue, 11 Aug 2026 18:23:11 +0300 Subject: Connectionists: FICN 2026: The IEEE co-sponsored International Conference on Future and Intelligent Communication and Networking, 8 - 10 December, 2026 | San Antonio, USA. Message-ID: [Apologies if you got multiple copies of this invitation] The International Conference on Future and Intelligent Communication and Networking (FICN 2026) https://emergingtechnet.org/FICN2026/index.php 8 - 10 December, 2026 | San Antonio, USA. Hybrid Event and Technically Co-Sponsored by IEEE Lone Star Section *AISSA 2026 CFP:* Future and intelligent communication and networking represent a rapidly evolving field driven by the integration of advanced technologies such as artificial intelligence, machine learning, and next-generation wireless systems like 5G and beyond. As global connectivity demands continue to grow, modern networks are becoming more autonomous, adaptive, and capable of handling massive volumes of data in real time. These intelligent systems enable innovations across various domains, including smart cities, IoT ecosystems, autonomous transportation, and cloud-edge infrastructures. The convergence of communication technologies with intelligent decision-making is transforming how networks are designed, managed, and optimized, paving the way for more efficient, secure, and scalable digital environments. The International Conference on Future and Intelligent Communication and Networking (FICN 2026) invites researchers, academics, industry professionals, and practitioners to submit original and high-quality research papers addressing the latest advancements in communication systems and intelligent networking technologies. FICN 2026 aims to provide a global platform for sharing innovative ideas, emerging trends, and practical solutions in the fields of next-generation communication, artificial intelligence-driven networks, and smart connectivity systems. The conference will bring together experts from academia, industry, and government to foster collaboration and advance the state of the art in intelligent communication technologies. We invite the submission of original papers on all topics related to Intelligent Systems with special interest in but not limited to: ? Future and intelligent communication and networking Systems ? Cloud, Edge Computing and Fog Networking ? Artificial Intelligence for Communication and Networking ? Generative AI, Foundation Models, and Large Language Models ? Computer Vision, Image and Video Processing ? Future Communication and Networking Applications ? 5G, 6G, and Beyond Communication Systems ? Computational Language and Human-Centered Systems ? Internet of Things (IoT) and Industrial IoT ? Communication and Networking Security ? Agentic AI communication and networking Systems ? Generative AI, Foundation Models, and Large Language Models for Communication and Networking Applications *Submissions Guidelines and Proceedings* Manuscripts should be prepared in 10-point font using the IEEE 8.5" x 11" two-column format. All papers should be in PDF format, and submitted electronically at Paper Submission Link. A full paper can be up to 8 pages (including all figures, tables and references). Length of short papers can be up to 6 pages. Submitted papers must present original unpublished research that is not currently under review for any other conference or journal. All submissions are peer-reviewed by at least three reviewers. Accepted papers will appear in the conference Proceeding, and to be submitted to IEEE Xplore for inclusion. *Important Dates:* - *Paper submission deadline: 10 September 2026* - Notification of acceptance: 1 November 2026 - Camera-ready Submission: 15 November 2026 *Contact:* Please send any inquiry on FICN to: emergingtechnetwork at gmail.com -------------- next part -------------- An HTML attachment was scrubbed... URL: From christine.Basta at alexu.edu.eg Wed Aug 12 05:41:25 2026 From: christine.Basta at alexu.edu.eg (christine.Basta) Date: Wed, 12 Aug 2026 09:41:25 +0000 Subject: Connectionists: Third and Last Call for Papers: The 7th Workshop on Gender Bias in NLP (GeBNLP 2026) Message-ID: Third and Last Call for Papers: The 7th Workshop on Gender Bias in NLP (GeBNLP 2026) 5th AACL & 15th IJCNLP, November 9, 2026 Hengqin, China (Remote-only mode) About the Workshop The GeBNLP workshop serves as a leading venue for the study, evaluation, and mitigation of gender bias in natural language processing. As large language models (LLMs) become foundational to recent NLP applications, addressing how these systems represent and affect different genders, alongside intersecting demographic axes such as race, ethnicity, nationality, religion, and ability, remains a critical challenge for the AI community. While foundational progress has been made in algorithmic debiasing and balanced data collection, recent large-scale evaluations reveal that state-of-the-art models continue to exhibit persistent stereotypes and confidence disparities across intersectional identities (Siddique et al., 2024; Savoldi et al., 2024). Our workshop serves as a multidisciplinary bridge, enabling researchers to define shared standards for tasks and metrics that ensure technical advancements are deeply rooted in the social and ethical realities of systemic harm (Dai et al., 2024). Topics of Interest We invite submissions on a wide range of topics related to gender bias in NLP, including but not limited to: * Measurement and Evaluation: New metrics and datasets for quantifying bias in LLMs, MT, and multimodal systems. * Mitigation Strategies: Technical approaches to debiasing (e.g., fine-tuning, adapter-based methods, or prompting strategies). * Multilingual and Cross-Cultural Perspectives: Bias in low-resource languages or non-Western cultural contexts. * Intersectionality: Research exploring how gender bias intersects with race, disability, age, or nationality. * Ethical and Legal Frameworks: Policy implications and the "Human-in-the-loop" role in auditing AI systems. Authors are encouraged to go beyond binary gender definitions and discuss how their work addresses the complexity of intersecting stereotypes and the diverse demographic contexts involved. Submission Guidelines * Long Papers are up to 8 pages and short Papers are up to 4 pages (excluding references and appendix). * Non-Archival Submissions: Authors may opt for non-archival submission, allowing work of standard conference quality to be presented without being published in the official proceedings. * Submission Link (Blind submission is required) Important Dates * Paper Submission Deadline: Wednesday, September 2, 2026 * Pre-reviewed (ARR) submission deadline: Tuesday, September 29, 2026 * Notification of Acceptance: Friday, October 2, 2026 * Camera-Ready Version Due: Monday, October 12, 2026 * Workshop Date: Monday, November 9, 2026 Paper Integrity Policy (Important) We endorse the EMNLP 2026 Paper Integrity Policy. Submissions that misuse AI, include hallucinated citations, present thinly sliced contributions, or are entirely AI-generated may be desk rejected. AI-assisted writing is permitted, provided authors remain responsible for the work. Organizers Giuseppe Attanasio, Instituto de Telecomunica??es, Lisbon Christine Basta, Alexandria University & HiTZ, University of the Basque Agnieszka Fale?ska, University of Stuttgart Vera Neplenbroek, University of Amsterdam Debora Nozza, Bocconi University Karolina Sta?czak, ETH AI Center, Zurich Marta R. Costa-juss?, FAIR, Meta Christian Hardmeier, IT university of Copenhagen References Dai, Y., Gu, H., Wang, Y. and Wang, X., 2024, November. Mitigate extrinsic social bias in pre-trained language models via continuous prompts adjustment. In Proceedings of the 2024 conference on empirical methods in natural language processing (pp. 11068-11083). Siddique, Z., Turner, L. and Anke, L.E., 2024, November. Who is better at math, jenny or jingzhen? uncovering stereotypes in large language models. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (pp. 18601-18619). Savoldi, B., Papi, S., Negri, M., Guerberof-Arenas, A. and Bentivogli, L., 2024, November. What the harm? quantifying the tangible impact of gender bias in machine translation with a human-centered study. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (pp. 18048-18076). On behalf of the organizers, Christine Basta -------------- next part -------------- An HTML attachment was scrubbed... URL: From Michael_Frank at brown.edu Wed Aug 12 08:00:34 2026 From: Michael_Frank at brown.edu (Michael J Frank) Date: Wed, 12 Aug 2026 08:00:34 -0400 Subject: Connectionists: Junior and Senior faculty positions at Brown University Message-ID: We have two job openings: applications due Sept 15. Assistant Professor in Cognitive and Psychological Sciences. The Department of Cognitive and Psychological Sciences (CoPsy) at Brown University invites applications for a tenure-track Assistant Professor position beginning July 1, 2027. Candidates' research must focus on one or more of the following research themes: (1) the interface between artificial intelligence and human cognition, (2) collective cognition and behavior, (3) mechanisms of mental and brain health. We are seeking candidates whose primary focus is on basic, computational, or experimental science rather than clinical intervention, assessment, or service delivery. Successful candidates will be working toward building an externally funded nationally recognized research program, be interested in interacting with colleagues from a wide range of disciplines and academic backgrounds, and be able to provide effective instruction and advising to a diverse group of graduate and undergraduate students. The CoPsy department has a highly interdisciplinary research environment in the study of mind, brain, and behavior and offers undergraduate curricular programs in Psychology, Cognitive Science, Cognitive Neuroscience, and Behavioral Decision Sciences. The Department is located in the heart of campus and is associated with many Centers and Initiatives at the University, including the Carney Institute for Brain Science, the Watson Institute for International and Public Affairs, the Data Science Initiative, and the NSF AI Research Institute on Interaction for AI Assistants (ARIA). The CoPsy department is a diverse and inclusive academic community. Candidates should address in their cover letter how they would contribute to the research and/or teaching missions of this diverse and inclusive community. QUALIFICATIONS All Ph.D. requirements must be completed before July 1, 2027. APPLICATION INSTRUCTIONS. The following documents must be submitted online as PDFs to https://apply.interfolio.com/190677: (1) Curriculum vitae; (2) reprints or preprints of three representative publications; (3), a statement of research (strictly no more than 2 pages), in which the candidate demonstrates (a) their fit with one or more of the position?s research themes, (b) a track record of excellence in research, and (c) a well-specified research plan that is likely to lead to extramural research funding; (4) a statement of teaching interests and philosophy (strictly no more than 1 page), in which the candidates demonstrate their ability and readiness to contribute to teaching and mentoring at both the undergraduate and graduate level; and (5) a statement highlighting the candidate?s contributions to service (e.g., diversity initiatives, open science; no more than 1 page). Applicants must indicate three references whom Interfolio will prompt to upload letters of reference. All applications must be received by September 15, 2026. Brown University provides equal opportunity and prohibits discrimination, harassment and retaliation based upon a person?s race, color, religion, sex, age, national or ethnic origin, disability, veteran status, sexual orientation, gender identity, gender expression, or any other characteristic protected under applicable law, in the administration of its policies, programs, and activities. The University recognizes and rewards individuals on the basis of qualifications and performance. The University maintains certain affirmative action programs in compliance with applicable law. Associate or Full Professor in Cognitive and Psychological Sciences. The Department of Cognitive and Psychological Sciences (CoPsy) at Brown University invites applications for a tenured Associate or Full Professor position beginning July 1, 2027. Candidates' research must focus on one or more of the following research themes: (1) the interface between artificial intelligence and human cognition, (2) collective cognition and behavior, (3) mechanisms of mental and brain health. We are seeking candidates whose primary focus is on basic, computational, or experimental science rather than clinical intervention, assessment, or service delivery. Successful candidates will have established an externally funded, internationally recognized research program, have an interest in interacting with colleagues from a wide range of disciplines and academic backgrounds, and have demonstrated effective instruction and advising to a diverse group of graduate and undergraduate students. The CoPsy department has a highly interdisciplinary research environment in the study of mind, brain, and behavior and offers undergraduate curricular programs in Psychology, Cognitive Science, Cognitive Neuroscience, and Behavioral Decision Sciences. The Department is located in the heart of campus and is associated with many Centers and Initiatives at the University, including the Carney Institute for Brain Science, the Watson Institute for International and Public Affairs, the Data Science Initiative, and the NSF AI Research Institute on Interaction for AI Assistants (ARIA). The CoPsy department is a diverse and inclusive academic community. Candidates should address in their cover letter how they would contribute to the research and/or teaching missions of this diverse and inclusive community. APPLICATION INSTRUCTIONS. The following documents must be submitted online as PDFs to https://apply.interfolio.com/190696: (1) Curriculum vitae; (2) reprints or preprints of three representative publications; (3) a statement of research (strictly no more than 3 pages), in which the candidate demonstrates (a) their fit with one or more of the position?s research themes, (b) an impressive track record of excellence in research, and (b) a well-specified future research plan that is likely to generate extramural research funding; (4) a statement of teaching (strictly no more than 2 pages), in which the candidate demonstrates their ability to contribute to teaching and mentoring at both the undergraduate and graduate level; and (5) a statement highlighting the candidate?s contributions to service (e.g., diversity initiatives, open science; no more than 1 page). Applicants must submit three to five names and contact information for professional references, who will be asked later to submit letters. All applications must be received by October 15, 2026. Brown University provides equal opportunity and prohibits discrimination, harassment and retaliation based upon a person?s race, color, religion, sex, age, national or ethnic origin, disability, veteran status, sexual orientation, gender identity, gender expression, or any other characteristic protected under applicable law, in the administration of its policies, programs, and activities. The University recognizes and rewards individuals on the basis of qualifications and performance. The University maintains certain affirmative action programs in compliance with applicable law. Michael J Frank, PhD | Edgar L. Marston Professor Director, Nancy G Zimmerman Center for Computational Brain Science Laboratory of Neural Computation and Cognition Brown University website -------------- next part -------------- An HTML attachment was scrubbed... URL: From walter.senn at unibe.ch Wed Aug 12 11:34:44 2026 From: walter.senn at unibe.ch (walter.senn at unibe.ch) Date: Wed, 12 Aug 2026 15:34:44 +0000 Subject: Connectionists: =?windows-1252?q?Postdoc_Position_in_Computationa?= =?windows-1252?q?l_Neuroscience_/_NeuroAI_=96_University_of_Bern=2C_Switz?= =?windows-1252?q?erland?= Message-ID: Dear Connectionists, A postdoctoral position in Computational Neuroscience is available at the University of Bern, Switzerland, at the Department of Physiology, under the supervision of Walter Senn. The Project. At the intersection of neuroscience and artificial intelligence, this position offers the opportunity to generalize our recent cortical model of multi-head self-attention toward in-context learning driven by biologically plausible neuronal error-circuits. Our model provides a mechanistic framework that unifies attention-based learning with gradient-based synaptic plasticity. You will collaborate directly with the team behind The Virtual Brain, applying these network models to study neurological and psychiatric disorders through the lens of disrupted sensory, motor, and cognitive self-attention. Furthermore, your work will integrate with our ongoing research to extend the Neuronal Least-Action Principle into long-term temporal processing and spike-based learning paradigms. Ultimately, you will be tackling a core challenge in the field: harnessing online learning in neuronal circuits to both decode human brain function and architect novel AI algorithms. Qualifications. A PhD in computational neuroscience, machine learning, physics, applied mathematics, or a related quantitative field. Strong publication track record and excellent mathematical skills. What We Offer - An interdisciplinary and international research environment at the frontier of neuroscience and AI, featuring next-door interactions with the Mihai Petrovici and Jean-Pascal Pfister labs, and genuine project ownership. - Active collaborations across artificial intelligence, experimental neuroscience, and neuromorphic engineering beyond University of Bern. - An internationally competitive Swiss salary, and a beautiful workplace in the UNESCO-listed capital of Switzerland, with nature and mountains within walking distance. - The position starts with a 1-year contract (flexible starting date, ideally around October 1), with the possibility of extensions. How to Apply. Please send a single PDF document containing your CV, a statement of research interests, a publication list, and the contact details of two referees to walter.senn at unibe.ch and federico.benitez at unibe.ch. Informal inquiries are welcome. Applications will be reviewed upon arrival, latest by September 6, 2026; early applications are highly encouraged. Best regards, Walter Department of Physiology, University of Bern -------------- next part -------------- An HTML attachment was scrubbed... URL: From tomg at princeton.edu Thu Aug 13 12:51:59 2026 From: tomg at princeton.edu (Tom Griffiths) Date: Thu, 13 Aug 2026 16:51:59 +0000 Subject: Connectionists: Resources for teaching cognitive science and AI Message-ID: <135D4570-E7A8-4A19-A636-D2DC39CD5934@princeton.edu> Dear colleagues, If you are preparing a course on cognitive science or AI for the Fall, I wanted to share some resources that might be useful. Over the last decade I have been doing oral history interviews with cognitive scientists and am releasing them in a podcast called the Cognition Project. Currently we have interviews posted with Jerome Bruner, Molly Potter, Susan Carey, Lila Gleitman, Noam Chomsky, Jean Berko Gleason, Tom Bever, Dan Slobin, Richard Atkinson, Gordon Bower, Roger Shepard, Rich Shiffrin, Elizabeth Loftus, Ewart Thomas, Phil Johnson-Laird, Donald Norman, Eleanor Rosch, and Douglas Hofstadter (and Dedre Gentner will be posted this week). The interviews so far have focused on the cognitive revolution, mathematical psychology, and the first cognitive science conferences, but we?re moving into a section on different approaches to AI and the origins of neural networks. The interviews are designed to be accessible to undergraduates so they can be assigned in classes where you cover the work of these researchers. Apple: https://podcasts.apple.com/us/podcast/the-cognition-project/id1872765981 Spotify: https://open.spotify.com/show/7rhwBGhEQCtO9cBguazFsq I also wrote a book for a general audience based on these interviews that introduces different approaches researchers have taken in trying to use mathematics to understand the mind, starting with the symbolic approach of Leibniz and Boole and then introducing neural networks and probability and statistics. The book, The Laws of Thought, is also designed to be assigned in undergraduate courses as a supplement to primary sources: it focuses on the stories of the people behind these ideas, putting the ideas in context. Here are some kinds words from readers: Tom Griffiths? The Laws of Thought is a magnificent piece of work. Very few scholars today bother with the history of contemporary concepts and methods, but Griffiths does so in a very entertaining and informative manner. And the crystal-clear explanations of the math were unprecedented in my experience. The book will be very useful and enlightening for students of cognitive science at all levels for generations to come. ? Michael Tomasello, James F. Bonk Distinguished Professor of Psychology and Neuroscience, Duke University For those of us who study how the brain gives rise to thought, Tom Griffiths has written something extraordinary: a clear-eyed account of the mathematical ideas that have come closest to capturing the mind. The Laws of Thought is more than an intellectual history ? it is a roadmap of the mind itself, and one that neuroscience has only begun to read. It sets a bold new agenda for our field. ? Edward Chang, Professor and Chair of Neurosurgery, University of California, San Francisco The Laws of Thought is a remarkable achievement: lucid enough for general readers, rich enough for scholars, and illuminating for anyone interested in intelligence, human or artificial. Griffiths offers an insight on every page and a revelation in every chapter. ? Tamar Gendler, Vincent J. Scully Professor of Philosophy, Yale University What makes this work exceptional is the synthesis. You move from Aristotle to Boole to Turing to modern transformers without ever feeling like you?ve left the same intellectual thread. ? Brian Keating, Chancellor?s Distinguished Professor of Physics, University of California, San Diego Publisher: https://us.macmillan.com/books/9781250358363/thelawsofthought/ Amazon: https://www.amazon.com/dp/1250358353 Best wishes, Tom. -- Tom Griffiths Director, Princeton Laboratory for Artificial Intelligence Henry R. Luce Professor of Information Technology, Consciousness and Culture Departments of Psychology and Computer Science Princeton University http://cocosci.princeton.edu/tom/ From thomas_serre at brown.edu Thu Aug 13 07:24:27 2026 From: thomas_serre at brown.edu (thomas_serre at brown.edu) Date: Thu, 13 Aug 2026 07:24:27 -0400 Subject: Connectionists: =?utf-8?q?Postdoctoral_Fellowships_in_Computation?= =?utf-8?q?al_Neuroscience_=E2=80=94_NIH_T32_Training_Program=2C_Br?= =?utf-8?q?own_University_=28review_begins_Sept_1=29?= Message-ID: The Nancy G. Zimmerman Center for Computational Brain Science at Brown University is recruiting postdoctoral fellows in computational neuroscience for its NIH T32 Training Program. Eligible research spans brain and cognitive modeling across multiple scales and levels of analysis, from biophysics to artificial intelligence, with potential connections to mental health and psychiatry. Fellows work with a primary trainer in a computational field and may add a second mentor in clinical psychiatry, working in areas such as neuroimaging, neurostimulation, and digital phenotyping. Trainers span applied mathematics, computer science, engineering, neuroscience, neurosurgery, and psychiatry. Fellows join a growing cohort of postdocs across the Zimmerman Center, our sister ICoN T32 in computational, cognitive and systems neuroscience, and ARIA, Brown's new $20M NSF institute on trustworthy AI assistants. The program includes core seminars in grant writing, responsible conduct of research, and rigor and reproducibility, alongside an individualized curriculum. Two to three fellows are appointed each year. Requirements: a PhD and/or MD completed before the start date, research experience in a computational field, and a record of publication. Per NIH T32 rules, applicants must be a US citizen or permanent resident at the time of appointment. Applications are submitted through Interfolio: a CV, a statement of research interests, up to three representative publications, and the names of three references. Review begins September 1, 2026. Candidates are encouraged to contact individual trainers in advance to explore shared interests and to mention this in their statement. The list of faculty trainers: https://carney.brown.edu/education-training/postdocs/nih-t32-training-program https://apply.interfolio.com/190111 Thomas Serre, Ph.D. Thomas J. Watson, Sr. Professor of Science Brown University ? Carney Institute for Brain Science ? Cog. & Psych. Sci. ? CS Faculty Director, Center for Computation & Visualization Assoc. Director, Zimmerman Center for Computational Brain Science +1 (401) 863-1148 ? thomas-serre.com ? x.com/tserre ? bsky: @thomasserre -------------- next part -------------- An HTML attachment was scrubbed... URL: From m.denker at fz-juelich.de Thu Aug 13 07:48:06 2026 From: m.denker at fz-juelich.de (Michael Denker) Date: Thu, 13 Aug 2026 13:48:06 +0200 Subject: Connectionists: PhD opportunity "Multi-Scale Comparisons of Patterns of Concerted Activity and Causality in Cortical Neural Activity" Message-ID: <20eda7e0-45e9-4899-a783-524235149b0f@fz-juelich.de> Dear all, I am pleased to announce a PhD opportunity in Computational Neuroscience at the J?lich Research Centre, Germany titled "Multi-Scale Comparisons of Patterns of Concerted Activity and Causality in Cortical Neural Activity." The project combines methods from computational neuroscience and electrophysiological data analysis. The position is embedded in the Helmholtz School for Data Science in Life, Earth and Energy (HDS-LEE), which offers a structured PhD programme with additional training in data science methods. For the full project description and application procedure, please see the job advertisement: https://recruiting.fz-juelich.de/jobposting/a4f654429eb21bf57d00cb15af3381b6f489368f0 Best regards, Michael Denker -- Dr. Michael Denker Team Leader Data Science in Electro- and Optophysiology Behavioral Neuroscience Institute for Advanced Simulation (IAS-6) Tel: +49 2461 61-9471 Computational and Systems Neuroscience Fax: +49 2461 61-9460 J?lich Research Centrewww.csn.fz-juelich.de 52425 J?lich, Germany --------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------- Forschungszentrum J?lich GmbH 52425 J?lich Sitz der Gesellschaft: J?lich Eingetragen im Handelsregister des Amtsgerichts D?ren Nr. HR B 3498 Vorsitzender des Aufsichtsrats: MinDir Stefan M?ller Gesch?ftsf?hrung: Prof. Dr. Astrid Lambrecht (Vorsitzende), Dr. Stephanie Bauer (stellvertretende Vorsitzende), Prof. Dr. Ir. Pieter Jansens, Prof. Dr. Laurens Kuipers --------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------- -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: smime.p7s Type: application/pkcs7-signature Size: 4338 bytes Desc: S/MIME Cryptographic Signature URL: From emergingtechnetwork.publicity at gmail.com Thu Aug 13 08:30:15 2026 From: emergingtechnetwork.publicity at gmail.com (=?UTF-8?Q?Gizem_G=C3=BCltekin_Varkonyi?=) Date: Thu, 13 Aug 2026 15:30:15 +0300 Subject: Connectionists: FLLM2026 (Co-Sponsored by IEEE): The 4th International Conference on Foundation and Large Language Models, 17 - 20 November 2026 | Barcelona, Spain Message-ID: [Apologies if you got multiple copies of this invitation] The 4th International Conference on Foundation and Large Language Models (FLLM 2026) https://fllm-conference.org/2026/index.php 17 - 20 November 2026 | Barcelona, Spain Hybrid Conference and Technically Co-Sponsored by IEEE Spain Section *FLLM 2026 CFP:* With the emergence of foundation models (FMs) and Large Language Models (LLMs) that are trained on large amounts of data at scale and adaptable to a wide range of downstream applications, Artificial intelligence is experiencing a paradigm revolution. BERT, T5, ChatGPT, GPT-4, Falcon 180B, Codex, DALL-E, Whisper, and CLIP are now the foundation for new applications ranging from computer vision to protein sequence study and from speech recognition to coding. Earlier models had a reputation of starting from scratch with each new challenge. The capacity to experiment with, examine, and comprehend the capabilities and potentials of next-generation FMs is critical to undertaking this research and guiding its path. Nevertheless, these models are currently inaccessible as the resources required to train these models are highly concentrated in industry, and even the assets (data, code) required to replicate their training are frequently not released due to their demand in the real-time industry. At the moment, mostly large tech companies such as OpenAI, Google, Facebook, and Baidu can afford to construct FMs and LLMS. Despite the expected widely publicized use of FMs and LLMS, we still lack a comprehensive knowledge of how they operate, why they underperform, and what they are even capable of because of their emerging global qualities. To deal with these problems, we believe that much critical research on FMs and LLMS would necessitate extensive multidisciplinary collaboration, given their essentially social and technical structure. The International Conference on Foundation and Large Language Models (FLLM) addresses the architectures, applications, challenges, approaches, and future directions. We invite the submission of original papers on all topics with special interest in but not limited to: - *Architectures and Systems* - Transformers and Attention - Bidirectional Encoding - Autoregressive Models - Massive GPU Systems - Prompt Engineering - Multimodal LLMs - Fine-tuning - *Challenges* - Hallucination - Cost of Creation and Training - Energy and Sustainability Issues - Integration - Safety and Trustworthiness - Interpretability - Fairness - Social Impact - * Future Directions* - Generative AI - Explainability and EXplainable AI - Retrieval Augmented Generation (RAG) - Federated Learning for FLLM - Large Language Models Fine-Tuning on Graphs - Data Augmentation - * Natural Language Processing Applications* - Generation - Summarization - Rewrite - Search - Question Answering - Language Comprehension and Complex Reasoning - Clustering and Classification - * Applications* - Natural Language Processing - Communication Systems - Security and Privacy - Image Processing and Computer Vision - Life Sciences - Financial Systems *Call for Special Tracks Papers:* - *Short papers:* (4-6 pages). - *Poster papers:* (1-2 Pages) - *Position Paper:* (2-4 pages) - *PhD Symposium:* (Up to 6 Pages) - *Demos :* (Up to 2 Pages) - *Systemization of Knowledge (SoK) or Focused Surveys:* (10-15 pages) All accepted papers will be published in the conference proceedings and submitted to IEEE Xplore for publication. *Journal Special Issue:* Selected high quality papers will be invited for special issue submission at the *Information Processing & Management (impact factor : 8.1)* https://www.sciencedirect.com/journal/information-processing-and-management *Submissions Guidelines and Proceedings* Manuscripts should be prepared in 10-point font using the IEEE 8.5" x 11" two-column format. All papers should be in PDF format, and submitted electronically at Paper Submission Link. A full paper can be up to 8 pages (including all figures, tables and references). Extra pages (up to 2 pages) can be purchased for a fee. Submitted papers must present original unpublished research that is not currently under review for any other conference or journal. Papers not following these guidelines may be rejected without review. Also submissions received after the due date, exceeding length limit, or not appropriately structured may also not be considered. Authors may contact the Program Chair for further information or clarification. All submissions are peer-reviewed by at least three reviewers. Accepted papers will appear in the FLLM Proceeding, and be published by the IEEE Computer Society Conference Publishing Services and be submitted to IEEE Xplore for inclusion. Submitted papers must include original work, and must not be under consideration for another conference or journal. Authors of accepted papers are expected to present their work at the conference. *Important Dates:* - *Paper submission deadline: August 21, 2026 (Extended)* - Notification of acceptance: October 1 , 2026 - Camera-ready Submission: October 15, 2026 *Contact:* Please send any inquiry on FLLM to: info@ fllm-conference.org -------------- next part -------------- An HTML attachment was scrubbed... URL: From mcruzcunha at gmail.com Sat Aug 15 17:52:06 2026 From: mcruzcunha at gmail.com (Maria Manuela Cruz Cunha) Date: Sat, 15 Aug 2026 22:52:06 +0100 Subject: Connectionists: =?utf-8?q?=5BCFP=5D_AIDA_2026_=E2=80=93_Internati?= =?utf-8?q?onal_Conference_on_AI_and_Data_Analytics_=E2=80=93_Seville_/_Hy?= =?utf-8?q?brid_conference_=E2=80=93_Deadline_15_September?= In-Reply-To: <7328b066-937f-4187-a850-837d4b4a7afd@ipca.pt> References: <7328b066-937f-4187-a850-837d4b4a7afd@ipca.pt> Message-ID: <5d1407b6-ba40-4d96-931a-a4afe16d6418@ipca.pt> Dear Colleagues, We are pleased to invite submissions to *AIDA 2026 ? 1st International Conference on Artificial Intelligence and Data Analytics*, taking place in *Seville, Spain, on 11?13 November 2026*, with hybrid participation available. AIDA 2026 welcomes original research on advances and applications in Artificial Intelligence and Data Analytics, including: * Artificial Intelligence * Machine Learning and Deep Learning * Generative AI and Large Language Models * Data Analytics * Natural Language Processing * Computer Vision * Explainable and Responsible AI * Data Science and Advanced Analytics * Intelligent Systems and Robotics * AI-enabled Digital Transformation * AI applications across industry and services *Submission deadline:*15 September 2026 *Conference dates:*11?13 November 2026 *Venue:*Barcel? Sevilla Renacimiento, Seville, Spain *Format:*Hybrid ? on-site and online participation Further information and submission details are available at: https://aida.scika.org We would appreciate it if you could circulate this Call for Papers among colleagues and researchers who may be interested. Best regards, *Manuela Cruz-Cunha* Polytechnic University of C?vado and Ave, Portugal AIDA 2026 -------------- next part -------------- An HTML attachment was scrubbed... URL: From ludovico.montalcini at gmail.com Fri Aug 14 10:51:25 2026 From: ludovico.montalcini at gmail.com (Ludovico Montalcini) Date: Fri, 14 Aug 2026 16:51:25 +0200 Subject: Connectionists: =?utf-8?q?3rd_Call_for_Participation=3A_Internati?= =?utf-8?q?onal_Artificial_Intelligence_Summer_School_=E2=80=93_IAI?= =?utf-8?q?SS_2026=2C_Sept_20-24=2C_Riva_del_Sole_Resort_=26_SPA_?= =?utf-8?q?=40_Tuscany_*_Registration=3A_by_August_23_*?= Message-ID: * Apologies for multiple copies. Please forward to anybody who might be interested, thanks! * IAISS2026, A Residential Summer School in Frontier AI IAISS 2026 is a full-immersion 5-day Summer School in Tuscany on cutting-edge advances in AI and Generative AI with lectures delivered by world-renowned experts. ? Riva del Sole Resort & SPA @ Tuscany, Italy, ? Sept 20-24, 2026 ? https://2026.iaiss.cc iaiss at icas.cc ? LECTURERS: Each Lecturer will hold up to four lectures on one or more research topics. Stefano V. Albrecht, Nanyang Technological University, Singapore Rafal Bogacz, University of Oxford, UK Giuseppe De Giacomo, University of Oxford, UK Christoph Feichtenhofer, Meta Superintelligence Labs, USA Sven Giesselbach, T-Systems International, Germany Jamie Hayes, Google DeepMind, UK Sumi Helal, University of Bologna, Italy Albert Q. Jiang, Mistral AI, Paris, France & University of Cambridge, UK Marine Le Morvan, INRIA Saclay, France Bruno Lepri, FBK, Italy Sebastian Lins Risi, Sakana AI, Japan & IT University of Copenhagen, Denmark George Paliouras, National Center for Scientific Research "Demokritos", Athens, Greece Panos Pardalos, University of Florida, USA ? https://2026.iaiss.cc/lecturers/ ? https://2026.iaiss.cc/lectures/ ? REGISTRATION: by August 23 (AoE) https://2026.iaiss.cc/registration/ Applicant *may* submit (during the Registration Process) a research abstract (max 2 pages; any format) for presentation. School directors and the scientific committee will review the abstracts and will recommend for poster and/or short talk. Abstract can be submitted by August 23 during the registration process. PAST EDITIONS: https://2026.iaiss.cc/previous-editions/ IAISS Norbert Wiener Best Presentation Award in Artificial Intelligence: https://2026.iaiss.cc/best-presentation-award/ VENUE: https://2026.iaiss.cc/venue/ Riva del Sole Resort & SPA Localit? Riva del Sole? Castiglione della Pescaia (Grosseto) CAP 58043? Tuscany? Italy p: +39-0564-928111 e: booking.events at rivadelsole.it w: https://www.rivadelsole.it/en/ ? CERTIFICATE & 8 ECTS: The International Artificial Intelligence Summer School ? IAISS 2026 is a full-immersion five-day Summer School at the Riva del Sole Resort & SPA (Castiglione della Pescaia ? Grosseto ? Tuscany, Italy) on cutting-edge advances in AI and Generative AI with lectures delivered by world-renowned experts. The Summer School provides a stimulating environment for PhD students, Post-Docs, junior academics (only up to assistant professors), early career researchers, and industry leaders (and highly motivated, promising and brilliant Master students / BSc students). Participants will also have the chance to present their results with talks, and to interact with their colleagues, in a convivial, professional and productive environment. PhD students, PostDocs, Industry Practitioners and Junior Academics (only up to assistant professors) will be typical profiles of the IAISS attendants. The Summer School will involve a total of 36?40 hours of lectures, according to the academic system the final achievement will be equivalent to 8 ECTS points for the PhD Students (and some strongly motivated Master Student ? BSc Student) attending the Summer School. Language: English. To participate in the IAISS 2026, all attendants must (1/2) register for the school (by August 23) and (2/2) book accommodation at the school venue, ?Riva del Sole Resort & SPA? (by August 23); all attendants must stay at the ?Riva del Sole Resort & SPA?. Once accommodation has been booked at ?Riva del Sole Resort & SPA?, the participant must send this information (including the Booking Number) to the IAISS organizing committee (iaiss at icas.cc). IAISS is a residential school, so all lecturers and participants must reside in the same Hotel (Riva del Sole Resort & SPA). No exceptions are made. For privacy reasons, the Hotel cannot match people. If you have someone to share the apartment with, please send to the Hotel ( booking.events at rivadelsole.it ) the name, surname and email address. Otherwise the solution is to book an apartment in single use. IAISS is organized as a non-profit scientific event. Please note that only IAISS registered participants can book a room (in hotel or apartment) at the Riva del Sole Resort & SPA with the accommodation form attached in the registration email and with the IAISS Discounted Rates. The Booking Office of the Riva del Sole Resort & SPA will verify the names of the participants and the corresponding registration number to confirm the booking. Anyone interested in participating in IAISS 2026 should register as soon as possible: ? https://2026.iaiss.cc/registration/ See you in Riva del Sole in September! IAISS 2026 Organizing Committees. https://2026.iaiss.cc iaiss at icas.cc https://www.linkedin.com/groups/19146028/ Obviously, this is only a Call for Participation, to have complete and updated information we recommend you access the relevant website: https://2026.iaiss.cc the News Section: https://2026.iaiss.cc/category/news/ and the FAQ: https://2026.iaiss.cc/faq/ * Apologies for multiple copies. Please forward to anybody who might be interested, thanks! * -------------- next part -------------- An HTML attachment was scrubbed... URL: From lees4 at ornl.gov Fri Aug 14 09:12:34 2026 From: lees4 at ornl.gov (Lee, Sangkeun (Matt)) Date: Fri, 14 Aug 2026 13:12:34 +0000 Subject: Connectionists: [CFP] BTSD 2026: AI-Assisted Scientific Discovery at IEEE BigData - Abstracts Due Oct. 1 Message-ID: CALL FOR PAPERS The 6th International Workshop on Big Data and AI-Assisted Tools, Methods, and Use Cases for Innovative Scientific Discovery (BTSD 2026) Held in conjunction with the 2026 IEEE International Conference on Big Data (IEEE BigData 2026) Workshop website: https://sites.google.com/view/btsd2026 BTSD 2026 invites original research on artificial intelligence, machine learning, and big-data technologies that accelerate innovative scientific discovery. Topics of interest include, but are not limited to: - Neural networks and machine learning for scientific discovery - Generative artificial intelligence and large language models for science - Autonomous and agentic artificial intelligence systems - Scientific foundation models, surrogate models, and digital twins - Learning from large-scale, multimodal, and heterogeneous scientific data - Human-artificial-intelligence collaboration and trustworthy AI - Scientific workflows, automation, provenance, and reproducibility - High-performance and distributed computing for AI-enabled science - Applications in materials, energy, climate, biology, healthcare, transportation, and other scientific domains IMPORTANT DATES Paper abstract submission due: October 1, 2026 Full workshop paper submission due: October 7, 2026 Submission and formatting instructions: https://sites.google.com/view/btsd2026 Please share this call with interested researchers, practitioners, and students. Best regards, Sangkeun (Matt) Lee Oak Ridge National Laboratory On behalf of the BTSD 2026 Organizing Committee https://sites.google.com/view/btsd2026 -------------- next part -------------- An HTML attachment was scrubbed... URL: From boubchir at ai.univ-paris8.fr Sun Aug 16 05:47:17 2026 From: boubchir at ai.univ-paris8.fr (Larbi Boubchir) Date: Sun, 16 Aug 2026 10:47:17 +0100 Subject: Connectionists: [CfP] [Extended deadline] The 4th International Workshop on Artificial Learning for Cyber Security - AICCSA 2026 In-Reply-To: <86f45c76-0567-42cd-bf01-6c97aea32a8c@univ-paris8.fr> References: <1bb7c980-a81f-466f-9d68-9a369601d693@ai.univ-paris8.fr> <86f45c76-0567-42cd-bf01-6c97aea32a8c@univ-paris8.fr> Message-ID: <93e14be6-dd52-4016-b1fe-8f3c0e49f493@ai.univ-paris8.fr> _[Apologies for multiple postings]_ The 4th international workshop on Artificial Learning for Cyber Security (ALCS 2026) will be held at the University of University of Sharjah, UAE, from 26 to 29 october 2026, in conjunction with the ACS/IEEE 23th International Conference on Computer Systems and Applications (AICCSA 2026). *Overview* Artificial Learning (ML) has demonstrated its ability to resolve several complex problems in various fields including cybersecurity, which protects critical systems and sensitive information against digital attacks. This workshop aims to explore the recent artificial learning algorithms and techniques that can be utilized to resolve complex problems in the cybersecurity field such as cyberattacks, and data protection and security. It is an opportunity to present and discuss the recent fundamental and applied research works, providing novel solutions to emerging challenges in cybersecurity and other related fields. The topics of interest include, but are not limited to, the following: - Advanced AL algorithms and methods for data analytics (deep learning, statistical learning, reinforcement learning, federated learning, ?) - AL for cyberattacks (ransomware attacks, spoofing attacks, IoT attacks, identity-based attacks, Cloud attacks, cryptocurrency attacks, ?) - AL for Healthcare cybersecurity (patient data security, device medical security, ?) - AL for Biometric recognition and security (authentication, identification, access control, anti-spoofing attacks, behavioral biometrics, forensic biometrics, ?) - AL for Blockchain (secure data sharing, fraud detection, ?) - AL for Cryptography (key generation, adversarial attacks detection, ?) - AL for Telecommunication (AI-powered automation, predictive analytics, 5G/6G network optimization, ?) - Generative AI in Cybersecurity (vulnerabilities prediction, recommendation system, ?) - AL for data protection and security - AL for predictive and adaptive security - Related applications Important Dates August 20, 2026 (11:59 pm CST): Due date for full workshop papers submission August 27, 2026 : Notification of paper acceptance to authors Sept. 5, 2026: Camera-ready of accepted papers Sept. 5, 2026: Registration deadline Oct 26-29, 2026: Workshop Online Submission Submit via EasyChair: https://easychair.org/conferences/?conf=aiccsa2026 Select: International workshop on Artificial Learning for Cyber Security Contact Prof. Larbi Boubchir /(Workshop Chair/) University of Paris 8, France E-mail: larbi.boubchir at univ-paris8.fr -------------- next part -------------- An HTML attachment was scrubbed... URL: From marysue0301 at gmail.com Sun Aug 16 11:29:22 2026 From: marysue0301 at gmail.com (Ruichen CONG) Date: Mon, 17 Aug 2026 00:29:22 +0900 Subject: Connectionists: CFP (Late Breaking Submission Open to Aug. 25, 2026): The 11th IEEE Cyber Science and Technology Congress (CyberSciTech 2026) - Melbourne, Australia, November 9-13, 2026 Message-ID: [Our apologies if you receive multiple copies of this CFP] Dear Colleagues, We are writing to invite you to submit your papers at The 11th IEEE Cyber Science and Technology Congress (CyberSciTech 2026) scheduled on November 9-13, 2026, Melbourne, Australia. https://cyber-science.org/2026/cyberscitech/ The Late Breaking Submission is quickly approaching on August 25, 2026. We look forward to your submissions! Submission link: https://edas.info/N35527 IMPORTANT DATES ----------------------------------------------- Regular Paper Submission Due: Aug. 15, 2026 (Extended) WiP/Poster/Wksp/SS Paper Due: Aug. 25, 2026 (Extended) Late Breaking Submission Due: Aug. 25, 2026 (Extended) Author Notification: Sep. 01, 2026 Paper Registration Due: Sep. 24, 2026 Camera-ready Submission Due: Oct. 01, 2026 CONGRESS INTRODUCTION -------------------------- Cyberspace, the seamless integration of physical, social, and mental spaces, is an integral part of our society, ranging from learning and entertainment to business and cultural activities, and so on. There are, however, in addition to its technical challenges, a number of pressing issues such as safety and trust associated with the cyberspace. To address these challenges, there is a need to establish new science and research portfolios that incorporate cyber-physical, cyber-social, cyber-intelligent, and cyber-life technologies in a cohesive and efficient manner. This is the aim of the IEEE Cyber Science and Technology Congress (CyberSciTech). IEEE CyberSciTech has been successfully held in Auckland, New Zealand, 2016, in Orlando, USA, 2017, in Athens, Greece, 2018, in Fukuoka, Japan, 2019, in Calgary, Canada, 2020 and 2021 (online due to COVID-19), in Calabria, Italy, 2022, in Abu Dhabi, UAE, 2023, in Boracay Island, Philippines, 2024, and in Hakodate City, Japan, 2025. In 2026, we will continue to offer IEEE CyberSciTech with the aim of providing a common platform for scientists, researchers, and engineers to share their latest ideas and advances in the broad scope of cyber-related science, technology, and application topics. In addition, this is also a platform to allow relevant stakeholders to get together, discuss and identify ongoing and emerging challenges, in order to understand and shape new cyber-enabled worlds. SCOPE AND TRACKSs ----------------------------------------------- Topics of interest include, but are not limited to: Regular Tracks (6-8 pages) Track 1: Cyberspace Theory & Technology - Cyberspace Property, Structure & Models - Cyber Pattern, Evolution, Ecology & Science - SDN/SDS, 5G/6G, Vehicle & Novel Network - Cloud, Fog, Edge & Green Computing - Big Data Analytics, Technology & Service - Infrastructures for Smart City/Country Track 2: Cyber Security, Privacy & Trust - Cyber Security, Safety & Resilience - Cyber Crime, Fraud, Abuse & Forensics - Cyber Attack, Terrorism, Warfare & Defense - Cyber Privacy, Trust & Insurance - Blockchain, DLT Techniques & Applications - Post-quantum Cryptography Track 3: Cyber Physical Computing & Systems - Cyber Physical Systems & Interfaces - Cyber Physical Dynamics & Disaster Relief - Cyber Manufacturing & Control - Embedded Systems & Software - Autonomous Robots & Vehicles - IoT, Digital Twin & Smart Systems Track 4: Cyber Social Computing & Networks - Social Networking & Computing - Computational Social Science - Crowd Sourcing, Sensing & Computing - Cyber Culture, Relation, Creation & Art - Cyber Social Right, Policy, Laws & Ethics - Cyber Learning, Economics & Politics Track 5: Cyber Intelligence & Cognitive Science - Cyber/Digital Brain & Artificial Intelligence - Hybrid & Hyper-connected Intelligence - Affective/Mind Cognition & Computing - Brain/Mind Machine Interface - AI Agents & Embodied Intelligence - Intelligent Object, Environment & Service Track 6: Cyber Life & Wellbeing - Cyber Life & Human Centric Computing - Cyber Medicine, Healthcare & Psychology - Cyborg/Wearable/Implantable Technology - Human/Animal Behavior Recognition - Personal Big Data & Personality Computing - Augmented/Mixed Reality & Metaverse IEEE CyberSciTech 2026 CALLS ---------------------------------- For original papers in: - Regular Tracks: 6-8 pages - WiP/Workshop/Special Session Tracks: 4-6 pages - Poster Track: 2 pages - LBI (Late Breaking Innovation) Track: 4-8 pages - All accepted conference, workshop, special session (SS), and poster papers will be published by IEEE in the Conference Proceedings (IEEE-DL and EI indexed). Selected high quality papers will be recommended to prestige journal special issues. Submission link: https://edas.info/N35527 SUBMISSION GUIDELINES ---------------------------------- Authors are invited to submit their original work that has not previously been submitted or published in any other venue. Regular, Work-in-Progress (WiP), Workshop/SS, Poster papers all need to be in IEEE CS format ( https://www.ieee.org/conferences/publishing/templates.html) CO-LOCATED CONFERENCES ---------------------------------- - The 24th IEEE International Conference on Pervasive Intelligence and Computing (PICom 2026) - The 24th IEEE International Conference on Dependable, Autonomic and Secure Computing (DASC 2026) - The 12th IEEE International Conference on Cloud and Big Data Computing (CBDCom 2026) Hosted by RMIT University, Australia -------------- next part -------------- An HTML attachment was scrubbed... URL: From alireza.modirshanechi at helmholtz-munich.de Sat Aug 15 08:55:42 2026 From: alireza.modirshanechi at helmholtz-munich.de (Alireza Modirshanechi) Date: Sat, 15 Aug 2026 12:55:42 +0000 Subject: Connectionists: =?iso-8859-1?q?Several_fully_funded_PhD_and_postd?= =?iso-8859-1?q?oc_positions_in_G=F6ttingen=2C_Germany?= Message-ID: Dear all, I am happy to announce that I will start as a junior professor and Emmy Noether awardee at the University of G?ttingen this October. I am recruiting several fully funded PhD students and postdocs to work at the intersection of cognitive science, computational neuroscience, and machine learning. For this hiring round, we are particularly interested in candidates who want to help us build a unified science of control in humans and machines. Projects may involve behavioral experiments, computational modeling, machine learning theory, or algorithm design. The positions are: * Fully funded at 100% TV-L E13 * Flexible in their start date, preferably before February 2027 * Open for applications on a rolling basis, with the first review round on September 15, 2026 We welcome candidates from cognitive science, psychology, neuroscience, computer science, mathematics, physics, electrical or computer engineering, and related fields. More information about the positions and how to apply: https://modirlab.github.io/open-positions.html More information about our research: https://modirlab.github.io/research.html Please feel free to share this announcement with anyone who may be interested. Best wishes, Alireza --- Dr. Alireza Modirshanechi Incoming Professor at the University of G?ttingen https://modirlab.github.io/ Helmholtz Zentrum M?nchen - Deutsches Forschungszentrum f?r Gesundheit und Umwelt (GmbH) Ingolst?dter Landstra?e 1, D-85764 Neuherberg, https://www.helmholtz-munich.de Gesch?ftsf?hrung: Prof. Dr. Dr. h.c. mult. Martin Hrabe de Angelis (komm.), Dr. Michael Frieser | Aufsichtsratsvorsitzender: MinDir Thomas Romes Registergericht: Amtsgericht M?nchen HRB 6466 | USt-IdNr. DE 129521671 -------------- next part -------------- An HTML attachment was scrubbed... URL: From caspar.schwiedrzik at googlemail.com Mon Aug 17 03:19:52 2026 From: caspar.schwiedrzik at googlemail.com (Caspar M. Schwiedrzik) Date: Mon, 17 Aug 2026 09:19:52 +0200 Subject: Connectionists: Deadline extended: Postdoctoral researcher in Neuropixels recordings, predictive processing, and navigation In-Reply-To: References: Message-ID: We are looking for an outstanding postdoctoral researcher to establish and lead a new research program using Neuropixels recordings in behaving mice to study predictive processing at the interface of vision, space, and navigation. We will investigate how the brain learns regularities in structured environments and uses them to generate predictions about upcoming events. The successful candidate will develop experiments in which mice navigate controlled environments while encountering visual objects or events with learnable statistical structure. The goal is to understand how neural populations represent expectations, prediction errors, and the interaction between visual and spatial information during navigation. This new research program will be part of the lab?s broader cross-species research program on predictive processing and experience-dependent plasticity. Related experiments are conducted in the lab in non-human primates and humans, using electrophysiology, fMRI, (intracranial) EEG, and behavioral approaches. The position therefore offers the opportunity to develop mechanistic experiments in rodents while contributing to an integrated research program within the lab across species, methods, and spatial scales. The lab seeks to understand the cortical basis and computational principles of perception and experience-dependent plasticity in the brain. See https://doi.org/10.64898/2025.11.30.691163, https://doi.org/10.1038/s41467-024-51543-y, and https://doi.org/10.64898/2025.12.17.694881 for recent examples of our work. The postdoc will play a key role in expanding these efforts toward high-density electrophysiology in behaving mice. Experiments will take place at the new research building THINK ? Center for Theoretical and Integrative Neuroscience and Cognitive Science at Ruhr University Bochum (https://www.ruhr-uni-bochum.de/think/index.html.en). THINK provides a highly interdisciplinary environment for integrative neuroscience and cognitive science, with dedicated infrastructure for human and animal research, neuroimaging, behavioral experimentation, data analysis, and high-performance computing. The postdoc will join a growing team in the lab, including researchers working on electrophysiology, fMRI, and behavior in mice. The position is funded through the Research Center One Health Ruhr of the University Alliance Ruhr. One Health Ruhr brings together researchers from neuroscience, molecular biology, water research, cancer research, and related fields to study fundamental mechanisms of health and disease across biological and environmental scales. The new application deadline is 31.08.2026. Please find the official job ad and further information here: https://jobs.ruhr-uni-bochum.de/jobposting/a6646cd83e1d4ed7e6d72b4ead787ec28f531ce90 From lzhang at cse.ust.hk Sun Aug 16 22:40:45 2026 From: lzhang at cse.ust.hk (lzhang at cse.ust.hk) Date: Mon, 17 Aug 2026 02:40:45 +0000 Subject: Connectionists: Postdoctoral Position: AI-Driven Peptide Drug Design and Modeling Message-ID: We are seeking a highly motivated Postdoctoral Researcher to join ourinternational team of experts on AI, peptide drug design, mRNA display, andimmunology. Based at the Department of Computer Science and Engineering (CSE),The Hong Kong University of Science and Technology (HKUST), you will lead thedevelopment and application of advanced computational methodologies?integratingartificial intelligence, molecular dynamics simulations, and moleculardocking?to accelerate cyclic peptide drug discovery. Supported by industry,your work will focus on algorithm design for modeling and designing noveltherapeutics. This includes establishing predictive models for both natural andnon-natural amino acid cyclic peptides, optimizing targets for receptor complexand membrane transport, and characterizing the essential parameters governingmembrane permeability for peptide delivery. Applicants should hold a Ph.D. in Computer Science, Bioinformatics,Computational Biology, or a closely related field, along with demonstratedexperience in training or fine-tuning foundation models for structural biology,a strong background in bioinformatics data curation, processing, and cleaning,and a solid track record of relevant academic publications. We offer a competitive salary commensurate with experience and an initial1-year contract with the possibility of renewal and long-term employment basedon performance and mutual agreement. Applications are accepted on a rolling basis and will be reviewedimmediately upon receipt until the position is filled. Please submit yourCV/resume, a brief cover letter outlining your relevant computational/AIexperience, and a list of publications to **drpahk at gmail.com mailto:drpahk at gmail.com **. -------------- next part -------------- An HTML attachment was scrubbed... URL: From George.Cybenko at dartmouth.edu Mon Aug 17 08:59:39 2026 From: George.Cybenko at dartmouth.edu (George Cybenko) Date: Mon, 17 Aug 2026 08:59:39 -0400 Subject: Connectionists: AI@50 videos... Message-ID: In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -------------- next part -------------- An HTML attachment was scrubbed... URL: From johan.suykens at esat.kuleuven.be Mon Aug 17 11:01:48 2026 From: johan.suykens at esat.kuleuven.be (Johan Suykens) Date: Mon, 17 Aug 2026 17:01:48 +0200 Subject: Connectionists: AI@50 videos... In-Reply-To: References: Message-ID: <10822652-f646-402e-a318-5d6017a39939@esat.kuleuven.be> very nice George, many thanks for sharing! Maybe a right moment to organize a follow-up AI at 70 meeting? Best regards, Johan On 8/17/2026 2:59 PM, George Cybenko wrote: > In 2006, we held a 50th anniversary conference at Dartmouth to commemorate > the historic 1956 meeting that coined the term "artificial > intelligence."? At that time, > five of the surviving original participants attended. Today they are > all gone. > > I just recently posted videos of the talks from the AI at 50 meeting at > > https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ > > INobody foresaw what is happening in AI today and in fact, > there were several digs?against neural networks etc. > > Enjoy! > > Regards > George Cybenko > -- > George Cybenko, Dorothy and Walter Gramm Professor of Engineering > Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: > gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: > Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -------------- next part -------------- An HTML attachment was scrubbed... URL: From terry at snl.salk.edu Mon Aug 17 13:30:39 2026 From: terry at snl.salk.edu (Terry Sejnowski) Date: Mon, 17 Aug 2026 10:30:39 -0700 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: Message-ID: Jean-Marc - I was there and gave a talk, the only one on neural networks.? I learned a lot from the other speakers. I just listened to the questions at the end of my talk.? I pretty much predicted what was going to happen in AI and thought it would take 50 years.? I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution.? Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list).? At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: > > In case you did not see this: Some amazing lectures and (now > historical) perspectives? the most interesting parts are when they > tried and predicted the future of AI! > > *From:*Connectionists > *On Behalf Of *George Cybenko > *Sent:* Monday, August 17, 2026 6:00 AM > *To:* connectionists at mailman.srv.cs.cmu.edu > *Subject:* Connectionists: AI at 50 videos... > > In 2006, we held a 50th anniversary conference at Dartmouth to commemorate > > the historic 1956 meeting that coined the term "artificial > intelligence."? At that time, > > five of the surviving original participants attended.? Today they are > all gone. > > I just recently posted videos of the talks from the AI at 50 meeting at > > https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ > > > INobody foresaw what is happening in AI today and in fact, > > there were several digs?against neural networks etc. > > Enjoy! > > Regards > > George Cybenko > > -- > George Cybenko, Dorothy and Walter Gramm Professor of Engineering > Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA > email: gvc at dartmouth.edu > FAX:??????????? 603 646-9024 > Cell:?????????? 603 369-1133 > Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -------------- next part -------------- An HTML attachment was scrubbed... URL: From Tanuj.Gulati at csmc.edu Mon Aug 17 18:00:58 2026 From: Tanuj.Gulati at csmc.edu (Gulati, Tanuj) Date: Mon, 17 Aug 2026 22:00:58 +0000 Subject: Connectionists: Postdoctoral Position - Cedars-Sinai, Los Angeles Message-ID: Postdoctoral Position - Electrophysiology | Cortico-cerebellar dynamics | Stroke Recovery Motor Systems Neuroscience and Neural Engineering Laboratory Cedars-Sinai Health Sciences University, Los Angeles We have a NIH-funded postdoctoral position available in the Motor Systems Neuroscience and Neural Engineering Laboratory at Cedars-Sinai's Center for Neural Science and Medicine in Los Angeles. Our lab uses a combination of systems neuroscience & neural engineering methods, along with causal manipulation of networks, to understand the process of motor skill consolidation in the brain and its breakdown after injury. A specific focus is on cortico-cerebellar dynamics and sleep's role in motor learning and rehabilitation (Rangwani et al, Cell Reports 2025; Abbasi et al, Science Advances 2024; Fleischer et al, eNeuro 2024, Gulati et al, Nat Neurosci 2014, 2017). We are also utilizing this knowledge to develop novel neurotechnology that promotes motor recovery after stroke (Abbasi et al, J Neural Engineering and Rehabilitation 2021, Simpson et al, Front Neurol 2023; Gulati et al, J Neurosci 2015). Building on this work, we are currently testing cerebellar neuromodulation approaches to promote stroke recovery, and recording across motor cortex and cerebellum in rodents to characterize spike and local field potential (LFP) dynamics as recovery proceeds. The postdoctoral fellow will be involved in multi-site electrophysiological recordings, closed-loop stimulation and brain-machine interface experiments in rodent models using neuropixels and TDT electrophysiology workstation, spike-LFP data analyses, and manuscript preparation. There will also be an opportunity to mentor junior trainees (interns/graduate students). Salary is negotiable. Requirements Applicants are expected to have a PhD in Biomedical Engineering, Neuroscience, or a related discipline. Strong candidates will have experience in one or more of the following: spike/LFP electrophysiology and analysis, data analysis and programming (MATLAB/Python), animal training, or experimental neurophysiology. Candidates with strong computational/analytical experience (e.g., spike sorting, LFP/spectral analysis, spike-field coherence) but more limited hands-on experimental experience will also be considered. Application Instructions Interested applicants should email me with their CV and contact information for 2 references. Email: tanuj.gulati at csmc.edu Lab website: www.gulatilab.org IMPORTANT WARNING: This message is intended for the use of the person or entity to which it is addressed and may contain information that is privileged and confidential, the disclosure of which is governed by applicable law. If the reader of this message is not the intended recipient, or the employee or agent responsible for delivering it to the intended recipient, you are hereby notified that any dissemination, distribution or copying of this information is strictly prohibited. Thank you for your cooperation. -------------- next part -------------- An HTML attachment was scrubbed... URL: From giovanni.stanco at unina.it Tue Aug 18 04:00:47 2026 From: giovanni.stanco at unina.it (GIOVANNI STANCO) Date: Tue, 18 Aug 2026 08:00:47 +0000 Subject: Connectionists: [CFP: IWNC'26 - CNSM'26] Extended Deadline - Third International Workshop on Integrated Wireless Networking and Computing 2026 In-Reply-To: References: Message-ID: The submission deadline for workshop papers is August 31, 2026 (Extended). Third International Workshop on Integrated Wireless Networking and Computing 2026 (IWNC'26) Joint with the 22nd International Conference on Network and Service Management (CNSM 2026) Alcal? de Henares (Madrid), Spain // 26 - 30 October, 2026 https://sites.google.com/view/iwnc-2026/ CALL FOR PAPERS The Third International Workshop on Integrated Wireless Networking and Computing 2026 (IWNC'26) invites high-quality submissions of papers describing original and unpublished research results regarding the use of intelligent computing capabilities at the edge or within the network. TOPICS OF INTEREST Topics of interest include, but are not limited to: ? AI/ML-driven optimization in wireless systems ? Management and orchestration architectures and techniques for next-gen networks ? Orchestration solutions for control plane, core, and RAN ? Joint design of communication and computation functions ? Programmable wireless networks and smart NICs ? In-Network Computing architectures and applications ? Cross-layer and cross-domain design approaches ? Early-stage research ideas and system prototypes on Integrated Wireless Networking and Computing ? Testbeds, emulators, and benchmarking for wireless networks ? Wireless networks for AI applications ? AI and ML in Edge Computing ? Advancements in wireless networks for Federated Learning ? Resilience, scalability, and adaptability of next-gen networks ? Cybersecurity Challenges in Integrated Wireless Networking and Computing ? Security and Privacy in Wireless Networks ? Energy-aware and Sustainable Wireless Networks ? O-RAN and MEC solutions ? O-RAN orchestration and management SUBMISSION GUIDELINES Submitted manuscripts should use IEEE 2-column conference style and are limited to 6 pages (including references). All papers accepted by a workshop will be published in the CNSM 2026 Proceedings and will be sent for inclusion in the IEEE eXplore digital library. The submission and revision process will be managed through the EDAS system (submission link: https://edas.info/N35574). IMPORTANT DATES ? Workshop papers submission deadline: August 31, 2026 ? Notification of acceptance: September 11, 2026 ? Camera-ready deadline: September 18, 2026 ? Workshop: October 30, 2026 We look forward to your valuable contributions and hope to see you at the workshop. Best regards, Stefania Zinno and Giovanni Stanco Third International Workshop on Integrated Wireless Networking and Computing 2026 (IWNC'26) -------------- next part -------------- An HTML attachment was scrubbed... URL: From cyndi.gundlach at hmhn.org Tue Aug 18 12:00:38 2026 From: cyndi.gundlach at hmhn.org (Cyndi Gundlach) Date: Tue, 18 Aug 2026 12:00:38 -0400 Subject: Connectionists: POST DOC FELLOW, EPILEPSY: (position available in New Jersey) Message-ID: Please post this job listing in the Connectionist 08.17.26 POST DOC FELLOW, EPILEPSY- connectionist (only PhD).docx POST DOC FELLOW, EPILEPSY: A postdoctoral research scientist with expertise in Neurosciences to assist our Neuroscience Faculty with their research efforts Education, Knowledge, Skills and Abilities Required: 1. Graduate of an accredited PhD program. 2. Prior training in Neurosciences, Signal Processing, AI or a related discipline. 1. The Research Epilepsy Fellows are encouraged to ideate, develop and complete a year-long research project related to EEG, epilepsy or brain networks. The trainees will work closely with a mentor and an interdisciplinary team consisting of Epileptologists, Neurologists, Neurosurgeons and researchers. The trainees will also participate in physician initiated and industry sponsored research. The trainees are encouraged to publish their findings in a research paper within a year of completing the fellowship 2. The trainee will shadow the Epileptologist in the Epilepsy monitoring unit and in the outpatient clinic. In addition, the trainees will monitor and review video EEG studies in the EMU under the supervision of an Epileptologist. Only licensed fellows may conduct patient evaluations. 3. The trainees will monitor and review routine EEG, rapid EEG, and continuous EEG under the supervision of an Epileptologist. The cases of patients with refractory seizures will be presented by the trainees in a weekly Epilepsy conference. The trainees will also review, discuss and present the intracranial EEG studies of patients who undergo monitoring with stereo EEG or subdural grids and independently form a hypothesis about the plausible seizure onset zone and the surgical plan. Only licensed fellows may provide EEG read reports for sign off by supervising epilepsy attendings. Non licensed providers will monitor studies to ensure continuous monitoring compliance and ensure quality of patient care through escalation of events to the on service neurophysiologist during identified seizure events. 4. The trainees will learn about Epilepsy through multiple formats including lectures, discussions and ward rounds (curriculum recommended by AAN). In addition, the trainees will participate in weekly multidisciplinary clinical conferences and case presentations. The trainees will also discuss interesting journal articles in monthly journal clubs. 5. Analysis and collecting data as well as organizing data in a useful format for publications. Responsible for publishing the study results and findings in peer-reviewed journals and for presentation at scientific meetings. Maintain research files and ensure participant confidentiality 6. Fellows are responsible for establishing research paradigms and designs for all proposed research interests. Responsible for all related grant writing to support the funding of the epilepsy program and EEG lab. Facilitate research collaboration with academicians inside and external to the Neuroscience Institute. Generation of comprehensive research databases to include all collected data to differentiate between diseases versus health participants. Creation and development of a database of epilepsy patients associated with seizure types and patient progression for work up in the EMU and surgical epilepsy program. To participate in ongoing continuing education programs and activities and to share acquired information with research faculty. 7. Mentor research students and research assistants as directed by the Laboratory Director. Participate in ongoing continuing education programs and activities and share acquired information with research faculty. 8. Other duties and/or projects as assigned. 9. Adheres to HMH Organizational competencies and standards of behavior. Please email cyndi.gundlach at hmhn.org your CV All my best, *Cyndi Gundlach* Fellowship Coordinator Neuroscience Center - Neurology Jersey Shore University Medical Center *PTO Alert: * 1945 Route 33 | Neptune, NJ 07753 T: 732-897-2250 | C: 732-832-9287 | F: E: cyndi.gundlach at hmhn.org This email and any files transmitted with it are confidential and are intended solely for the use of the individual or entity to which they are addressed. This communication may contain material protected by the attorney-client privilege. If you are not the intended recipient or the person responsible for delivering it to the addressee, please be advised that you have received this communication in error and that any dissemination, forwarding, printing or copying of this communication is prohibited. If you have received this message in error, please contact me at the above address or number. -------------- next part -------------- An HTML attachment was scrubbed... URL: From evomusart at gmail.com Tue Aug 18 06:57:11 2026 From: evomusart at gmail.com (EvoMUSART) Date: Tue, 18 Aug 2026 11:57:11 +0100 Subject: Connectionists: Call for Papers: 16th International Conference on Artificial Intelligence in Music, Sound, Art and Design (EvoMusArt 2027) Message-ID: Call for papers for the 16th International Conference on Artificial Intelligence in Music, Sound, Art and Design (EvoMusArt) ? Please distribute ? Apologies for cross-posting ------------------------------------------------ The 16th International Conference on Artificial Intelligence in Music, Sound, Art and Design (EvoMusArt) will take place on 31 March ? 2 April, 2027, as part of the evostar event. *EvoMusArt webpage: *https://www.evostar.org/2027/evomusart/ *Submission deadline: *1 November 2026 *Conference: *31 March ? 2 April 2027 EvoMusArt is a multidisciplinary conference that brings together researchers working on the application of Artificial Neural Networks, Evolutionary Computation, Swarm Intelligence, Cellular Automata, Artificial Life (Alife), Generative AI, Foundation Models, and other Artificial Intelligence (AI) techniques in creative and artistic fields such as Visual Art, Music, Architecture, Video, Digital Games, Poetry, Design, and Interactive Media. The conference provides a forum for presenting and discussing novel research, artistic practices, systems, and applications that explore computational creativity, human-AI co-creation, and emerging creative technologies. Submissions must be at most 14 pages long, excluding references, in Springer Lecture Notes in Computer Science (LNCS) format. Each submission must be anonymised for a double-blind review process. Accepted papers will be presented orally or as posters at the event and included in the EvoMusArt proceedings published by Springer Nature in a dedicated volume of the LNCS series. Submissions should address the use of AI techniques (e.g. Evolutionary Computation, Artificial Neural Networks, Alife, Machine Learning (ML), Deep Learning, and Swarm Intelligence) in creative and artistic domains, including art, music, design, architecture, and related fields. Topics of interest span generation, automation, computer-aided creativity and creativity support tools, and theoretical aspects of computational creativity, including but not limited to: ? Systems that create drawings, images, animations, videos, sculptures, poetry, text, designs, webpages, buildings, virtual environments, and other creative artefacts; ? Systems that create musical pieces, sounds, instruments, voices, sound effects, soundscapes, and multimodal artistic experiences; ? Systems that create artefacts such as game content, architecture, furniture, industrial products, and digital experiences based on aesthetic and functional criteria; ? Generative AI, foundation models, large language models (LLMs), multimodal models, and diffusion-based creative systems; ? Computational creativity in digital games, interactive storytelling, and narrative generation; ? Virtual Reality (VR), Augmented Reality (AR), Mixed Reality (MR), and Extended Reality (XR) for artistic and creative applications, including immersive art installations and interactive creative environments; ? Robotic-based Evolutionary Art and Music; ? Embodied AI, tangible and wearable interfaces, and smart environments for creative applications; ? Digital fabrication, 3D printing, and computational approaches to physical artefact creation; ? Other artificial, generative, evolutionary, or biologically inspired techniques in the fields of Computer Music, Computer Art, Design, and Digital Culture; ? Techniques for automatic fitness assignment; ? AI agents and autonomous creative systems; ? Systems in which analysis or interpretation of artworks is combined with AI techniques to generate novel artefacts; ? Systems that use AI for the analysis, understanding, restoration, preservation, or curation of artistic and cultural heritage resources; ? Human-AI co-creation, mixed-initiative systems, intelligent interfaces, and AI-based creativity support tools; ? New ways of integrating users and audiences into the evolutionary, generative, or creative cycle; ? User-centred evaluation of creative AI systems and creative experiences; ? Analysis and evaluation of the artistic potential of biologically inspired and AI-based creative systems, their creative processes, and resulting artefacts; ? Collaborative, distributed, and multi-user creative systems, networked art environments, and collective computational creativity; ? Contextualisation of Creative AI in cultural, economic, social, political, legal, ethical, ecological, and sustainability-related discourse. ? Computational Aesthetics, Experimental Aesthetics, Emotional Response, Engagement, Surprise, Novelty, and Meaning; ? Representation techniques and creative knowledge representations; ? Explainability, transparency, authorship, ownership, and ethics in Creative AI; ? Surveys of the current state of the art; identification of strengths and weaknesses; comparative analyses and taxonomies; ? Validation and evaluation methodologies for creative systems and generated artefacts; ? Studies on the applicability of these techniques to related domains; ? New models designed to promote creativity through evolutionary computation, Alife, ML, and hybrid approaches. More information on the submission process and topics: https://www.evostar.org/2027/evomusart/ Papers published in previous editions of EvoMusArt: https://evomusart-index.dei.uc.pt We look forward to seeing you at EvoMUSART 2027 in Mainz! The EvoMUSART 2027 organisers S?rgio M. Rebelo Patrick Donnelly Nereida Rodr?guez-Fern?ndez (publication chair) -------------- next part -------------- An HTML attachment was scrubbed... URL: From augenstein at di.ku.dk Tue Aug 18 07:47:20 2026 From: augenstein at di.ku.dk (Isabelle Augenstein) Date: Tue, 18 Aug 2026 11:47:20 +0000 Subject: Connectionists: Postdoc in Mechanistic Understanding of AI Reasoning at University of Copenhagen Message-ID: <29EDD97A-2ACB-47FD-8DA0-83C69901931C@ku.dk> We are seeking to recruit a postdoctoral researcher to work on understanding the mechanisms of AI reasoning. Their core duty will be to advance our understanding of how LLMs acquire, integrate and reason over evolving, pluralistic knowledge using mechanistic interpretability methods. This can involve the development of novel conceptual frameworks, methods, or empirical studies. The postdoc?s duties will also to a small degree include teaching or supervision. The position is offered in the context of the context of an ERC Starting Grant held by Isabelle Augenstein on ?Explainable and Robust Automatic Fact Checking (ExplainYourself)?, as well the Pioneer Centre for AI. The position will be offered from 1 February 2027 or as close to this date as possible, for a period of two years, with the possibility of extension contingent on funding availability. Apply here by 30 September 2026, 23:59 CET to be considered: https://candidate.hr-manager.net/ApplicationInit.aspx/?cid=3010&departmentId=19976&ProjectId=167176&MediaId=4917&SkipAdvertisement=false Who are we looking for? Applicants should hold a PhD degree or equivalent in Computer Science or a related field, and have good written and oral English skills. The assessment of qualifications will also be made based on previous scientific publications and relevant work experience. The ideal candidate would have an educational background, prior research or work experience on Natural Language Processing, Machine Learning, and/or Explainable AI. Our group and research -- and what do we offer? The successful candidate will join the CopeNLU group at the University of Copenhagen. CopeNLU is a vibrant and collaborative research group led by Isabelle Augenstein and Pepa Atanasova with a focus on fair and accountable NLP. We are interested in core methodology research on interpretability, explainability and bias detection; as well as applications to tasks such as fact checking and cross-cultural learning. With a strong focus on both foundational and applied research, we provide a platform for exploring cutting-edge topics in NLP, while also emphasising the importance of transparent and responsible AI development. We are affiliated with the Pioneer Centre for AI at the Department of Computer Science, Faculty of SCIENCE, University of Copenhagen, located in central Copenhagen. The Pioneer Centre focuses on fundamental AI research, and within an interdisciplinary framework, develops platforms, methods, and practices that address society?s greatest challenges. It consists of seven AI research themes, with one being Speech and Language. The Natural Language Processing research environment at the University of Copenhagen is internationally leading, as e.g. evidenced by it being ranked second in Europe according to CSRankings. Isabelle Augenstein, Dr. Scient., Ph.D. Professor and Deputy Head of Department for Research Department of Computer Science University of Copenhagen ?stervold Observatory ?ster Voldgade 3 1350 Copenhagen augenstein at di.ku.dk http://isabelleaugenstein.github.io/ -------------- next part -------------- An HTML attachment was scrubbed... URL: From swickbrennan at yahoo.com Tue Aug 18 21:00:16 2026 From: swickbrennan at yahoo.com (Brennan Swick) Date: Wed, 19 Aug 2026 01:00:16 +0000 (UTC) Subject: Connectionists: [CFP] Final Call for Papers - NeuRo-SymBolic World Models (RoBoWoMo) @ IROS 2026 References: <685833180.3631121.1787101216912.ref@mail.yahoo.com> Message-ID: <685833180.3631121.1787101216912@mail.yahoo.com> TL;DR: Event: NeuRo-SymBolic World Models (RoBoWoMo) Workshop at IROS 2026 (Pittsburgh, US) Date: September 27, 2026 Website: https://worldmodelworkshop.github.io/ Submission Deadline: August 25, 2026 OpenReview submission link: https://bit.ly/SubmitToRoBoWoMo? We are organizing the NeuRo-SymBolic World Models (RoBoWoMo) Workshop at IROS 2026 (Sept 27 in Pittsburgh) and would love for you to submit! Call for Papers:? - Half Papers: Up to 4 pages (excluding references/appendices) - Full Papers: Up to 8 pages (excluding references/appendices) We welcome submissions covering neuro-symbolic unification, hybrid architectures, task-driven world modeling, benchmarking/evaluation, etc. All accepted papers will have the opportunity to be presented via lightning talks and a dedicated poster session, with the strongest submissions selected for longer spotlight presentations. For more details on formatting, submission policies, and topics of interest, please visit our Call for Papers page: https://worldmodelworkshop.github.io/call_for_papers/? ? Important Dates (AoE): - Submission Deadline: August 25, 2026 - Author Notification: September 10, 2026 - Workshop Date: September 27, 2026 For schedule details: https://worldmodelworkshop.github.io/schedule? ? Invited Speakers and Panelists:? - Sherry Yang (New York University & Google DeepMind) - Yilun Du (Harvard University) - Tom Silver & Yixuan Huang (Princeton University) - Siddharth Srivastava (Arizona State University) - Emre Ugur (Bogazici University) - Sungjin Ahn (KAIST & New York University) - Jiajun Wu (Stanford University) -------------- next part -------------- An HTML attachment was scrubbed... URL: From frothga at sandia.gov Tue Aug 18 17:21:52 2026 From: frothga at sandia.gov (Rothganger, Fred) Date: Tue, 18 Aug 2026 21:21:52 +0000 Subject: Connectionists: AI@50 videos... In-Reply-To: References: Message-ID: I predict that if we ever encounter an alien intelligence, it will be non-biological (or rather, formerly-biological). This is based on the notion that machine intelligence is the normal end-state of evolution, sort of like Marovec's Mind Children. ________________________________ From: Connectionists on behalf of Terry Sejnowski Sent: Monday, August 17, 2026 11:30 AM To: director at inc.ucsd.edu ; Fellous, Jean-Marc ; connectionists at mailman.srv.cs.cmu.edu Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... You don't often get email from terry at snl.salk.edu. Learn why this is important Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -------------- next part -------------- An HTML attachment was scrubbed... URL: From fxcosta at ucp.pt Wed Aug 19 04:00:00 2026 From: fxcosta at ucp.pt (Felipe Xavier Costa) Date: Wed, 19 Aug 2026 08:00:00 +0000 Subject: Connectionists: =?utf-8?q?2_Weeks_Left_=E2=80=93_Submit_to_COMPLE?= =?utf-8?q?X_NETWORKS_2026_Now!?= Message-ID: Dear colleagues, ?? Time is running out! Submit your research to COMPLEX NETWORKS 2026 by September 2 and join an exciting interdisciplinary gathering of scientists working on complex systems, social networks, infrastructure, and more. ? https://complexnetworks.org/submission/ Choose between: * Full Papers (12 pages) ? Springer Proceedings * Extended Abstracts (4 pages) ? Book of Abstracts (with ISBN) ? Selected contributions may be invited for journal publication in: * PLOS Complex Systems * Applied Network Science * Entropy, and more ? Keynote Speakers: Alexandra Brintrup, Nitesh Chawla, Kimmo Kaski, Bruno Lepri, Eckehard Scholl ? Tutorials: December 1 | ? Conference: December 2?4 in Granada, Spain See you in Granada ? On behalf of COMPLEX NETWORKS 2026 Organizing Committee, Felipe Xavier Costa, PhD Postdoctoral Researcher https://fxcosta-phd.github.io/ Universidade Cat?lica Portuguesa www.cbr.fm.ucp.pt -------------- next part -------------- An HTML attachment was scrubbed... URL: From btissam.errahmadi at gmail.com Wed Aug 19 06:19:09 2026 From: btissam.errahmadi at gmail.com (Btissam Er-Rahmadi) Date: Wed, 19 Aug 2026 11:19:09 +0100 Subject: Connectionists: 3rd CFP of the 3rd NORA Workshop @AACL-IJCNLP 2026 Message-ID: Hi all, Apologies for cross-posting. This is the final CFP for the workshop. *Call for Papers* --------------------------------------------------------------------------------- *3rd* Workshop on KNOwledge GRaphs & Agentic Systems Interplay (*NORA*) 10 November 2026, Hengqin, China Web: https://nora-workshop.github.io/AACL2026/ LinkedIn: https://www.linkedin.com/company/nora-knowledge-graphs-agentic-systems-interplay --------------------------------------------------------------------------------- In conjunction with *AACL-IJCNLP 2026, November 6-10* --------------------------------------------------------------------------------- *Workshop Overview* --------------------------------------------------------------------------------- Agents have experienced significant growth in recent years, largely due to the rapid technological advancements of Large Language Models (LLMs). Although these agents benefit from LLMs? advanced generation proficiency, they still suffer from catastrophic forgetting and a limited context window size compared to the agents? needs in terms of contextual information. Knowledge Graphs (KGs) are a powerful paradigm for structuring and managing connected pieces of information while unlocking deeper insights than traditional methods. Their value is immense for tasks that require context, integration, inter-linking, and reasoning. However, this power comes at the cost of significant upfront and ongoing investment in construction, curation, and specialised expertise. The NORA workshop aims at analysing and discussing emerging and novel practices, ongoing research efforts and validated or deployed innovative solutions that showcase the growing synergy between LLMs agents and KGs. --------------------------------------------------------------------------------- *Topics of Interest* --------------------------------------------------------------------------------- We welcome submissions and participation from intradisciplinary, interdisciplinary and multidisciplinary researchers and industry & public sector practitioners in the areas of Knowledge Graphs, Knowledge Engineering and Reasoning, Advanced NLP, GenAI, and AI Agents. We especially welcome contributions that provide theoretical insights, propose new approaches, or introduce new grounded solutions in real-world applications such as enterprise, smart assistance & chat, healthcare, finance, tourism, etc. We invite submissions in this non-exhaustive list of topics of interest, including, but not limited to: - Agents for Complex Reasoning over KGs - Agents and KGs for private and proactive personal assistants & Personalisation - Architectures for Persistent Agent Memory - Benchmarking Agent Memory Performance - Collaborative & Shared Agent Memories - Context Engineering enhanced by KGs - Domain-Specific Memory Architectures - From Unstructured Experience to Structured and Graph-Based Memory - KGs serving agents' memories: Episodic (experiences, events, etc.), Semantic (facts, concepts, etc.), and Procedural (skills, tasks, etc.) - Memory Grounding & Hallucination Mitigation - Memory Indexing & Retrieval for Agents - Multi-Lingual & Multi-modal integrations - Personalization vs. Generalization in Memory ------------------------------------------------------------------------------------ *Important Dates Anywhere on Earth (AoE)* ------------------------------------------------------------------------------------ - Regular Submissions Deadline: September 9th, 2026 - ARR Commitment Deadline: September 14th, 2026 - Notification of Acceptance: October 1, 2026 - Camera-Ready Papers Due: October 12, 2026 - Workshop date: November 10th, 2026 ------------------------------------------------------------------------------------ Submission Guidelines, Policies, and Awards are available on the website. ------------------------------------------------------------------------------------ --------------------------------------------------------------------------------- *Organization* --------------------------------------------------------------------------------- - Btissam Er-Rahmadi, Independent Researcher, UK - Sebastien Montella, Huawei Technologies R&D UK Ltd, UK - Damien Graux, EcoVadis, UK - Andre Melo, Huawei Technologies R&D UK Ltd, UK - Hajira Jabeen, University Hospital Cologne, Germany Best, Btissam Er-Rahmadi -------------- next part -------------- An HTML attachment was scrubbed... URL: From jose at rubic.rutgers.edu Wed Aug 19 10:24:42 2026 From: jose at rubic.rutgers.edu (=?utf-8?B?U3RlcGhlbiBKb3PDqSBIYW5zb24=?=) Date: Wed, 19 Aug 2026 14:24:42 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: Message-ID: Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- [cid:part1.XnvX3hEx.pO10Zvgy at rubic.rutgers.edu] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.jpg Type: image/jpeg Size: 57383 bytes Desc: signature.jpg URL: From arbib at usc.edu Wed Aug 19 14:39:00 2026 From: arbib at usc.edu (Michael Arbib) Date: Wed, 19 Aug 2026 18:39:00 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: Message-ID: An added historical note. The same activity that led to the Dartmouth meeting also led to the book Shannon, C. E., & McCarthy, J. (1956). Automata studies. Princeton, NJ: Princeton University Press. It contains classic papers on finite automata, Turing machines, and (on a more speculative note, with British authors) synthesis of automata. Worth a look even now, but with a very different emphasis on GOFAI. I suppose Turing is the bridge. ________________________________ From: Connectionists on behalf of Stephen Jos? Hanson Sent: Wednesday, August 19, 2026 7:24 AM To: Terry Sejnowski ; director at inc.ucsd.edu ; Fellous, Jean-Marc ; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political.? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- [cid:part1.XnvX3hEx.pO10Zvgy at rubic.rutgers.edu] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.jpg Type: image/jpeg Size: 57383 bytes Desc: signature.jpg URL: From jose at rubic.rutgers.edu Wed Aug 19 14:48:58 2026 From: jose at rubic.rutgers.edu (=?utf-8?B?U3RlcGhlbiBKb3PDqSBIYW5zb24=?=) Date: Wed, 19 Aug 2026 18:48:58 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: Message-ID: Yes, I agree. Von Neumann was also at the MACY meetings and of course developing CA with Ulam in the 40s.. The ghost of cybernetics is everywhere. On 8/19/26 14:39, Michael Arbib wrote: An added historical note. The same activity that led to the Dartmouth meeting also led to the book Shannon, C. E., & McCarthy, J. (1956). Automata studies. Princeton, NJ: Princeton University Press. It contains classic papers on finite automata, Turing machines, and (on a more speculative note, with British authors) synthesis of automata. Worth a look even now, but with a very different emphasis on GOFAI. I suppose Turing is the bridge. ________________________________ From: Connectionists on behalf of Stephen Jos? Hanson Sent: Wednesday, August 19, 2026 7:24 AM To: Terry Sejnowski ; director at inc.ucsd.edu ; Fellous, Jean-Marc ; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political.? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- [cid:part1.jLBr4HO8.6Ht31CoB at rubic.rutgers.edu] -- [cid:part1.jLBr4HO8.6Ht31CoB at rubic.rutgers.edu] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.jpg Type: image/jpeg Size: 57383 bytes Desc: signature.jpg URL: From jose at rubic.rutgers.edu Wed Aug 19 14:28:31 2026 From: jose at rubic.rutgers.edu (=?utf-8?B?U3RlcGhlbiBKb3PDqSBIYW5zb24=?=) Date: Wed, 19 Aug 2026 18:28:31 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: oh gary please, not even close. Explain the kernel of an LLM please. btw you;ve been on so many sides of this over the last 3 years... its hard to tell what kind of polygon you really are. Cheers, Stephen On 8/19/26 14:22, Gary Marcus wrote: yep: neural networks ascended, pushed symbols out of the way, and then symbols came back relabeled as harnesses, loops code interpreters, and tools. neurosymbolic AI, the hybrid of the two approaches, clearly beat either of the two alternatives on their own. claude code, biggest recent advance, shows that pretty clearly, when you look under the hood. gary On Aug 19, 2026, at 10:16?AM, Stephen Jos? Hanson wrote: ? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- -- [cid:part1.kPgDT0gO.TrYqL9w3 at rubic.rutgers.edu] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.jpg Type: image/jpeg Size: 57383 bytes Desc: signature.jpg URL: From gary.marcus at nyu.edu Wed Aug 19 14:38:17 2026 From: gary.marcus at nyu.edu (Gary Marcus) Date: Wed, 19 Aug 2026 11:38:17 -0700 Subject: Connectionists: FW: AI@50 videos... Message-ID: An HTML attachment was scrubbed... URL: From gary.marcus at nyu.edu Wed Aug 19 14:22:44 2026 From: gary.marcus at nyu.edu (Gary Marcus) Date: Wed, 19 Aug 2026 11:22:44 -0700 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: Message-ID: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> An HTML attachment was scrubbed... URL: From George.Cybenko at dartmouth.edu Wed Aug 19 14:39:03 2026 From: George.Cybenko at dartmouth.edu (George Cybenko) Date: Wed, 19 Aug 2026 11:39:03 -0700 Subject: Connectionists: AI@50 videos... In-Reply-To: <10822652-f646-402e-a318-5d6017a39939@esat.kuleuven.be> References: <10822652-f646-402e-a318-5d6017a39939@esat.kuleuven.be> Message-ID: Thanks, Johan. There is an AI at 70 but it is not technical. This was Dartmouth leadership?s call. See https://ai.dartmouth.edu/ai-70-what-must-remain-human Terry, I have your slides which I will post but the videos were selected by James More, not me, and this was all that could be burned onto 8 DVDs. Regards George -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu On Tue, Aug 18, 2026 at 1:03?AM Johan Suykens < johan.suykens at esat.kuleuven.be> wrote: > very nice George, many thanks for sharing! > > Maybe a right moment to organize a follow-up AI at 70 meeting? > > Best regards, > > Johan > > > On 8/17/2026 2:59 PM, George Cybenko wrote: > > In 2006, we held a 50th anniversary conference at Dartmouth to commemorate > the historic 1956 meeting that coined the term "artificial intelligence." > At that time, > five of the surviving original participants attended. Today they are all > gone. > > I just recently posted videos of the talks from the AI at 50 meeting at > > https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ > > INobody foresaw what is happening in AI today and in fact, > there were several digs against neural networks etc. > > Enjoy! > > Regards > George Cybenko > > -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering > Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA > > email: gvc at dartmouth.edu > FAX: 603 646-9024 > Cell: 603 369-1133 > Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu > > -------------- next part -------------- An HTML attachment was scrubbed... URL: From tgd at oregonstate.edu Wed Aug 19 17:52:04 2026 From: tgd at oregonstate.edu (Dietterich, Thomas) Date: Wed, 19 Aug 2026 21:52:04 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: Message-ID: I heard a story that when the International Joint Conference on Artificial Intelligence was being created, a decision ?was made? to exclude pattern recognition from the list of topics, and this resulted in the separation between the PAMI community and the rest of AI. Can anyone confirm whether this is true? It is certainly consistent with McCarthy?s goals. If so, it wasn?t specifically neural network models that were excluded, but pretty much anything with real (or floating point) numbers, probabilities, etc. --Tom Thomas G. Dietterich, University Distinguished Professor (Emeritus) School of EECS, Oregon State University US Mail: 1148 Kelley Engineering Center, Corvallis, OR 97331-5501 USA Office: 2063 Kelley Engineering Center Voice: 541-737-5559; FAX: 541-737-1300 https://web.engr.oregonstate.edu/~tgd/ From: Connectionists On Behalf Of Michael Arbib Sent: Wednesday, August 19, 2026 11:39 AM To: Terry Sejnowski ; director at inc.ucsd.edu; Fellous, Jean-Marc ; connectionists at mailman.srv.cs.cmu.edu; Stephen Jos? Hanson Subject: Re: Connectionists: FW: AI at 50 videos... [This email originated from outside of OSU. Use caution with links and attachments.] An added historical note. The same activity that led to the Dartmouth meeting also led to the book Shannon, C. E., & McCarthy, J. (1956). Automata studies. Princeton, NJ: Princeton University Press. It contains classic papers on finite automata, Turing machines, and (on a more speculative note, with British authors) synthesis of automata. Worth a look even now, but with a very different emphasis on GOFAI. I suppose Turing is the bridge. ________________________________ From: Connectionists > on behalf of Stephen Jos? Hanson > Sent: Wednesday, August 19, 2026 7:24 AM To: Terry Sejnowski >; director at inc.ucsd.edu >; Fellous, Jean-Marc >; connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: FW: AI at 50 videos... Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political.? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- [cid:image001.jpg at 01DD2FE6.9FD0FCC0] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image001.jpg Type: image/jpeg Size: 57383 bytes Desc: image001.jpg URL: From ASIM.ROY at asu.edu Wed Aug 19 23:01:22 2026 From: ASIM.ROY at asu.edu (Asim Roy) Date: Thu, 20 Aug 2026 03:01:22 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: The Horace Barlow note was after a furious private debate with some eminent neuro and cognitive scientists regarding Jennifer Aniston cells being evidence for grandmother cells. And this was at age 93 when he visited me after defending his theory. At that stage of his career, his insights were profound. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Asim Roy Sent: Wednesday, August 19, 2026 6:28 PM To: 'Gary Marcus' ; Stephen Jos? Hanson Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: FW: AI at 50 videos... I had some communication with Marvin Minsky and Jerry Fodor years ago. Here are two quotes from them: 1. Marvin Minsky: "Don?t pay any attention to the critics. Don?t even ignore them.? -By Sam Goldwyn 2. Jerry Fodor: ?Arguing with Connectionists is like arguing with zombies; both are dead, but neither has noticed it.? And here?s a note from a neuroscience pioneer Horace Barlow, Darwin?s great-grandson (Horace Barlow - Wikipedia), winner of many neuroscience prizes. Without his pioneering work poking into frog brains with microelectrodes, we would NOT have AI as we see it today. The grandmother cell theory essentially points to where cognition exists in an abstract form. And there is plenty of neurophysiological evidence for such single cell abstractions, starting with line orientation cells. ====================================================================================================================================== Dear Asim Yes, I would like to join your group, though I would (again) like to make it clear that the grandmother cell theory is not mine, though I still hold that something conforming amazingly well to what was conceptualised by Jerry Lettvin 50 years ago, really does exist! I am still actively interested in the topic, and have just sent off the final proofs of an essay bearing on it. I shall send you a copy of the final proof as soon as I am allowed to circulate it, and of course the more widely discussed it is the better pleased I shall be, though I fear that what I have written will not be universally accepted, at least at first! Best regards Horace ============================================================================================================= Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Connectionists > On Behalf Of Gary Marcus Sent: Wednesday, August 19, 2026 11:23 AM To: Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... yep: neural networks ascended, pushed symbols out of the way, and then symbols came back relabeled as harnesses, loops code interpreters, and tools. neurosymbolic AI, the hybrid of the two approaches, clearly beat either of the two alternatives on their own. claude code, biggest recent advance, shows that pretty clearly, when you look under the hood. gary On Aug 19, 2026, at 10:16?AM, Stephen Jos? Hanson > wrote: ? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- -------------- next part -------------- An HTML attachment was scrubbed... URL: From ASIM.ROY at asu.edu Wed Aug 19 21:28:13 2026 From: ASIM.ROY at asu.edu (Asim Roy) Date: Thu, 20 Aug 2026 01:28:13 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: I had some communication with Marvin Minsky and Jerry Fodor years ago. Here are two quotes from them: 1. Marvin Minsky: "Don?t pay any attention to the critics. Don?t even ignore them.? -By Sam Goldwyn 2. Jerry Fodor: ?Arguing with Connectionists is like arguing with zombies; both are dead, but neither has noticed it.? And here?s a note from a neuroscience pioneer Horace Barlow, Darwin?s great-grandson (Horace Barlow - Wikipedia), winner of many neuroscience prizes. Without his pioneering work poking into frog brains with microelectrodes, we would NOT have AI as we see it today. The grandmother cell theory essentially points to where cognition exists in an abstract form. And there is plenty of neurophysiological evidence for such single cell abstractions, starting with line orientation cells. ====================================================================================================================================== Dear Asim Yes, I would like to join your group, though I would (again) like to make it clear that the grandmother cell theory is not mine, though I still hold that something conforming amazingly well to what was conceptualised by Jerry Lettvin 50 years ago, really does exist! I am still actively interested in the topic, and have just sent off the final proofs of an essay bearing on it. I shall send you a copy of the final proof as soon as I am allowed to circulate it, and of course the more widely discussed it is the better pleased I shall be, though I fear that what I have written will not be universally accepted, at least at first! Best regards Horace ============================================================================================================= Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Connectionists On Behalf Of Gary Marcus Sent: Wednesday, August 19, 2026 11:23 AM To: Stephen Jos? Hanson Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... yep: neural networks ascended, pushed symbols out of the way, and then symbols came back relabeled as harnesses, loops code interpreters, and tools. neurosymbolic AI, the hybrid of the two approaches, clearly beat either of the two alternatives on their own. claude code, biggest recent advance, shows that pretty clearly, when you look under the hood. gary On Aug 19, 2026, at 10:16?AM, Stephen Jos? Hanson > wrote: ? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- -------------- next part -------------- An HTML attachment was scrubbed... URL: From minaiaa at gmail.com Thu Aug 20 01:51:27 2026 From: minaiaa at gmail.com (Ali Minai) Date: Thu, 20 Aug 2026 01:51:27 -0400 Subject: Connectionists: AI@50 videos... In-Reply-To: References: Message-ID: What a treasure! Thank you for sharing this. *Ali A. Minai, Ph.D.* Professor Complex Adaptive Systems Lab Department of Electrical & Computer Engineering 828 Rhodes Hall University of Cincinnati Cincinnati, OH 45221-0030 Phone: (513) 556-4783 Fax: (513) 556-7326 Email: Ali.Minai at uc.edu minaiaa at gmail.com WWW: https://researchdirectory.uc.edu/p/minaiaa On Mon, Aug 17, 2026 at 9:40?AM George Cybenko wrote: > In 2006, we held a 50th anniversary conference at Dartmouth to commemorate > the historic 1956 meeting that coined the term "artificial intelligence." > At that time, > five of the surviving original participants attended. Today they are all > gone. > > I just recently posted videos of the talks from the AI at 50 meeting at > > https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ > > INobody foresaw what is happening in AI today and in fact, > there were several digs against neural networks etc. > > Enjoy! > > Regards > George Cybenko > > -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering > Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA > > email: gvc at dartmouth.edu > FAX: 603 646-9024 > Cell: 603 369-1133 > Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu > > -------------- next part -------------- An HTML attachment was scrubbed... URL: From juergen.schmidhuber at kaust.edu.sa Thu Aug 20 05:41:12 2026 From: juergen.schmidhuber at kaust.edu.sa (Juergen Schmidhuber) Date: Thu, 20 Aug 2026 09:41:12 +0000 Subject: Connectionists: [BULK] [EXTERNAL] Re: FW: AI@50 videos... In-Reply-To: References: Message-ID: <9FB4521E-7FCE-4CC3-AD7D-3CE7AA53B60F@kaust.edu.sa> Indeed, there was AI long before the 1956 AI meeting, and it was often called cybernetics. Fun fact: cybernetics was closer to modern AI than what the 1956 AI guys proposed. In 1914, the Spaniard Leonardo Torres y Quevedo became the 20th century's first AI pioneer when he built the first working chess end game player (back then chess was considered as an activity restricted to the realms of intelligent creatures). Quevedo?s machine was still considered impressive decades later when another AI pioneer?Norbert Wiener [WI48]?played against it at the 1951 Paris conference on calculating machines and human thought, now often viewed as the first conference on AI [AI51][BRO21][BRU4]. ?One reason for inventing the term [AI] was to escape association with cybernetics,? as McCarthy once bluntly explained. ?I wished to avoid having either to accept Norbert Wiener as a guru or having to argue with him.? Source: https://caseorganic.substack.com/p/inside-the-very-human-origin-of-the Stephen also mentions "the connectionist revival in the 1980s (sorry Juergen).? This revival was at best a rehash of the deep learning revolution that started in Ukraine in 1965. For example, Ivakhnenko's 1971 paper [DEEP2] (in English) described a deep learning network with 8 layers. Application: predict the next token! Where the tokens described the British economy. References taken from: [DLH] J.S. Annotated History of Modern AI and Deep Learning. Technical Report IDSIA-22-22, IDSIA, Switzerland, 2022, updated 2025. Preprint arXiv:2212.11279. https://people.idsia.ch/~juergen/deep-learning-history.html Cheers, Juergen > On 19. Aug 2026, at 16:24, Stephen Jos? Hanson wrote: > > You don't often get email from jose at rubic.rutgers.edu. Learn why this is important > Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. > Well at least we know how that turned out. > Stephen > > On 8/17/26 13:30, Terry Sejnowski wrote: >> Jean-Marc - >> >> I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. >> >> I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. >> >> One of the questions at the end was about intelligent aliens who appear and don't have human brains. >> This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. >> >> I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. >> >> Terry >> >> ------ >> >> On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: >>> In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! >>> From: Connectionists On Behalf Of George Cybenko >>> Sent: Monday, August 17, 2026 6:00 AM >>> To: connectionists at mailman.srv.cs.cmu.edu >>> Subject: Connectionists: AI at 50 videos... >>> In 2006, we held a 50th anniversary conference at Dartmouth to commemorate >>> the historic 1956 meeting that coined the term "artificial intelligence." At that time, >>> five of the surviving original participants attended. Today they are all gone. >>> I just recently posted videos of the talks from the AI at 50 meeting at >>> https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ >>> INobody foresaw what is happening in AI today and in fact, >>> there were several digs against neural networks etc. >>> Enjoy! >>> Regards >>> George Cybenko >>> -- >>> George Cybenko, Dorothy and Walter Gramm Professor of Engineering >>> Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA >>> >>> email: gvc at dartmouth.edu >>> FAX: 603 646-9024 >>> Cell: 603 369-1133 >>> Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu >> > -- > From robert.kentridge at durham.ac.uk Thu Aug 20 05:53:07 2026 From: robert.kentridge at durham.ac.uk (KENTRIDGE, ROBERT W.) Date: Thu, 20 Aug 2026 09:53:07 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: It is worth pointing out that in Jerry Lettvin?s original paper on the grandmother cell the concept was presented as a joke. Lettvin certainly believed in the notion that one might identify cells in the frog?s brain as ?bug detectors?, but he appeared to be sceptical about the idea that such cells might exist in the human brain representing complex concepts like ?my grandmother?. There are a couple of great papers on the history of the grandmother cell: Ann-Sophie Barwich (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 Charles C Gross (2002) Genealogy of the "grandmother cell?. Neuroscientist 8(5):512-8. doi: 10.1177/107385802237175. I remember chatting to Charlie Gross about this around that time (he also reminisced about his part-time job mowing BF Skinner?s lawn). People often assume that the Quiroga et al (2005 Nature, 435. 1102 - 1107) paper on the Jennifer Aniston cell was positive evidence for the existence of grandmother cells in the human brain. This is far from the truth. The Jennifer Aniston cell does not only respond to Jennifer Aniston, but also to related concepts like ?Lisa Kudow? (another actress in ?Friends?), so activity in that cell does not uniquely code ?Jennifer Aniston? and cannot be used to read out the presence of Jennifer Aniston. Quiroga et al make this clear in the discussion of their findings. Bob Kentridge Psychology Department University of Durham Canadian Institute for Advanced Research Programme on Brain, Mind and Consciousness Sent from Outlook for Mac From: Connectionists on behalf of Asim Roy Date: Thursday, 20 August 2026 at 06:51 To: Gary Marcus ; Stephen Jos? Hanson Cc: director at inc.ucsd.edu ; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] The Horace Barlow note was after a furious private debate with some eminent neuro and cognitive scientists regarding Jennifer Aniston cells being evidence for grandmother cells. And this was at age 93 when he visited me after defending his theory. At that stage of his career, his insights were profound. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Asim Roy Sent: Wednesday, August 19, 2026 6:28 PM To: 'Gary Marcus' ; Stephen Jos? Hanson Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: FW: AI at 50 videos... I had some communication with Marvin Minsky and Jerry Fodor years ago. Here are two quotes from them: 1. Marvin Minsky: "Don?t pay any attention to the critics. Don?t even ignore them.? -By Sam Goldwyn 2. Jerry Fodor: ?Arguing with Connectionists is like arguing with zombies; both are dead, but neither has noticed it.? And here?s a note from a neuroscience pioneer Horace Barlow, Darwin?s great-grandson (Horace Barlow - Wikipedia), winner of many neuroscience prizes. Without his pioneering work poking into frog brains with microelectrodes, we would NOT have AI as we see it today. The grandmother cell theory essentially points to where cognition exists in an abstract form. And there is plenty of neurophysiological evidence for such single cell abstractions, starting with line orientation cells. ====================================================================================================================================== Dear Asim Yes, I would like to join your group, though I would (again) like to make it clear that the grandmother cell theory is not mine, though I still hold that something conforming amazingly well to what was conceptualised by Jerry Lettvin 50 years ago, really does exist! I am still actively interested in the topic, and have just sent off the final proofs of an essay bearing on it. I shall send you a copy of the final proof as soon as I am allowed to circulate it, and of course the more widely discussed it is the better pleased I shall be, though I fear that what I have written will not be universally accepted, at least at first! Best regards Horace ============================================================================================================= Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Connectionists > On Behalf Of Gary Marcus Sent: Wednesday, August 19, 2026 11:23 AM To: Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... yep: neural networks ascended, pushed symbols out of the way, and then symbols came back relabeled as harnesses, loops code interpreters, and tools. neurosymbolic AI, the hybrid of the two approaches, clearly beat either of the two alternatives on their own. claude code, biggest recent advance, shows that pretty clearly, when you look under the hood. gary On Aug 19, 2026, at 10:16?AM, Stephen Jos? Hanson > wrote: ? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- -------------- next part -------------- An HTML attachment was scrubbed... URL: From school at utia.cas.cz Thu Aug 20 08:12:29 2026 From: school at utia.cas.cz (Miroslav Karny) Date: Thu, 20 Aug 2026 14:12:29 +0200 Subject: Connectionists: [BULK] [EXTERNAL] Re: FW: AI@50 videos... In-Reply-To: <9FB4521E-7FCE-4CC3-AD7D-3CE7AA53B60F@kaust.edu.sa> References: <9FB4521E-7FCE-4CC3-AD7D-3CE7AA53B60F@kaust.edu.sa> Message-ID: <1787227934108151139@utia.cas.cz> ?Dear all, just for fun and a bit of thinking on AI. Almost surely you know that the term ?robot? appeared in R.U.R. a famous 1920 science fiction play by Karel ?apek. It is less known that in 1922 he published a science fiction novel The Absolute at Large (Tov?rna na absolutno in Czech) https://en.wikisource.org/wiki/The_Absolute_at_Large. I believe that it is highly relevant for the research community (especially chapter 14, if you map the word ?tacks? on LLMs research outputs ? ) Best regards Miroslav K?rn? https://utia.cas.cz/en/people/?pid=109 Dne ?t, 08/20/2026 12:22 odp., Juergen Schmidhuber napsal(a): > Indeed, there was AI long before the 1956 AI meeting, and it was often called cybernetics. > > Fun fact: cybernetics was closer to modern AI than what the 1956 AI guys proposed. > > In 1914, the Spaniard Leonardo Torres y Quevedo became the 20th century's first AI pioneer when he built the first working chess end game player (back then chess was considered as an activity restricted to the realms of intelligent creatures). > > Quevedo?s machine was still considered impressive decades later when another AI pioneer?Norbert Wiener [WI48]?played against it at the 1951 Paris conference on calculating machines and human thought, now often viewed as the first conference on AI [AI51][BRO21][BRU4]. > > ?One reason for inventing the term [AI] was to escape association with cybernetics,? as McCarthy once bluntly explained. ?I wished to avoid having either to accept Norbert Wiener as a guru or having to argue with him.? Source: https://caseorganic.substack.com/p/inside-the-very-human-origin-of-the > > Stephen also mentions "the connectionist revival in the 1980s (sorry Juergen).? > > This revival was at best a rehash of the deep learning revolution that started in Ukraine in 1965. > > For example, Ivakhnenko's 1971 paper [DEEP2] (in English) described a deep learning network with 8 layers. Application: predict the next token! Where the tokens described the British economy. > > > > References taken from: > > [DLH] J.S. Annotated History of Modern AI and Deep Learning. Technical Report IDSIA-22-22, IDSIA, Switzerland, 2022, updated 2025. Preprint arXiv:2212.11279. https://people.idsia.ch/~juergen/deep-learning-history.html > > > Cheers, > Juergen > > > > > > > On 19. Aug 2026, at 16:24, Stephen Jos? Hanson wrote: > > > > You don't often get email from jose at rubic.rutgers.edu. Learn why this is important > > Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. > > Well at least we know how that turned out. > > Stephen > > > > On 8/17/26 13:30, Terry Sejnowski wrote: > >> Jean-Marc - > >> > >> I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. > >> > >> I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. > >> > >> One of the questions at the end was about intelligent aliens who appear and don't have human brains. > >> This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. > >> > >> I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. > >> > >> Terry > >> > >> ------ > >> > >> On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: > >>> In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! > >>> From: Connectionists On Behalf Of George Cybenko > >>> Sent: Monday, August 17, 2026 6:00 AM > >>> To: connectionists at mailman.srv.cs.cmu.edu > >>> Subject: Connectionists: AI at 50 videos... > >>> In 2006, we held a 50th anniversary conference at Dartmouth to commemorate > >>> the historic 1956 meeting that coined the term "artificial intelligence." At that time, > >>> five of the surviving original participants attended. Today they are all gone. > >>> I just recently posted videos of the talks from the AI at 50 meeting at > >>> https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ > >>> INobody foresaw what is happening in AI today and in fact, > >>> there were several digs against neural networks etc. > >>> Enjoy! > >>> Regards > >>> George Cybenko > >>> -- > >>> George Cybenko, Dorothy and Walter Gramm Professor of Engineering > >>> Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA > >>> > >>> email: gvc at dartmouth.edu > >>> FAX: 603 646-9024 > >>> Cell: 603 369-1133 > >>> Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu > >> > > -- > > > > > -------------- next part -------------- An HTML attachment was scrubbed... URL: From icaitemspisa at gmail.com Thu Aug 20 06:35:35 2026 From: icaitemspisa at gmail.com (ICAI TEMS 2026) Date: Thu, 20 Aug 2026 12:35:35 +0200 Subject: Connectionists: =?utf-8?q?=5BCFP=5D_IEEE_ICAI-TEMS_2026_=E2=80=93?= =?utf-8?q?_Final_Extension=3A_Full_Papers_due_25_August_2026_=28strict=29?= Message-ID: <348fe450-3fc8-4b7d-b1c5-d9c2d06164aa@gmail.com> Dear Colleagues, Due to several requests received in the last few days, the Organizing Committee of IEEE ICAI-TEMS 2026 has granted a final extension for full paper submission: *NEW DEADLINE: 25 August 2026 ? STRICT, no further extensions will be granted.* IEEE ICAI-TEMS 2026 ? 2nd International Conference on Application of Information Technologies in Engineering, Management and Science Pisa, Italy ? 10?13 November 2026 Sponsored by the IEEE Technology and Engineering Management Society (TEMS) ABOUT ICAI-TEMS is an interdisciplinary platform for original research on the concepts, theories and applications of cutting-edge information technologies across engineering, technology management and science. The conference fosters dialogue between academia and industry, and includes a dedicated industry forum bringing together corporate leaders, academics and students. TRACKS INCLUDE ? Artificial Intelligence Methods for Diagnostics ? Computer Vision ? Security and Trust of Systems and Applications ? Intelligent and Cognitive Robotic Systems ? CAD-centric Digital Manufacturing ? Internal Representations and Mechanistic Interpretability in Deep Learning ? AECO Industry & AI ? Technological Innovations and AI for Decentralized Energy Management ? Digital Technologies, Process Analytics and AI for Healthcare Management ? Decision Intelligence, Smart Agriculture, Wearable Sensing, Sustainability, Urban Digital Tools, Intelligent Transport Systems, and more Full list of tracks and topics: https://www.icai-tems.eu/call_for_paper.html#tracks KEY DATES ?*Full paper submission: 25 August 2026 (extended, strict)* ? Notification of acceptance: 11 September 2026 ? Camera-ready submission: 25 September 2026 ? Conference: 10?13 November 2026 PUBLICATION Accepted and presented papers will be published in the IEEE ICAI-TEMS 2026 conference proceedings and submitted for inclusion in IEEE Xplore and Scopus. SUBMISSION Author guidelines and submission link: https://www.icai-tems.eu/call_for_paper.html For more informations visit the website www.icai-tems.eu We look forward to welcoming you in Pisa. Best regards, Pietro Calabrese, Publicity Chair, IEEE ICAI-TEMS 2026 -------------- next part -------------- An HTML attachment was scrubbed... URL: From j.v.stone at sheffield.ac.uk Thu Aug 20 07:08:38 2026 From: j.v.stone at sheffield.ac.uk (James V Stone) Date: Thu, 20 Aug 2026 12:08:38 +0100 Subject: Connectionists: Fwd: AI@50 videos... References: Message-ID: <6AB663E1-3522-4A0B-9F5E-C957CFE98582@sheffield.ac.uk> Some of you may be interested in this recent book: The Artificial Intelligence Papers: Original Research Papers With Tutorial Commentaries by JV Stone Details here: https://jamesstone.sites.sheffield.ac.uk/books/theaipapers? James V Stone - TheAIPapers sites.sheffield.ac.uk -- James V Stone Visiting Professor, Sheffield University, UK. Web: https://jamesstone.sites.sheffield.ac.uk/books Begin forwarded message: From: George Cybenko Subject: Connectionists: AI at 50 videos... Date: 17 August 2026 at 13:59:39 BST To: connectionists at mailman.srv.cs.cmu.edu In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu <> -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: AG8ngQWGbMZ3Fjr4WQkpsSEyxh-mKXmu6XFzy0eWDMNQEOZTzzlbqRlyufkey-sc5g3mcnI599tWvKjAsICMCaFpkkLQPV5e4uOegBiX8y2Lko23rBEw_tHz0ueytMQbV63_ZhoQjBexL3g2jtfLP8q4JW8kRvQs-YXJ9sz6B2L5-iNWIdutoyf8rafa8CrG=w16383.png Type: image/png Size: 199539 bytes Desc: not available URL: From r.t.radulescu at uu.nl Thu Aug 20 10:56:56 2026 From: r.t.radulescu at uu.nl (Radulescu, R.T. (Roxana)) Date: Thu, 20 Aug 2026 14:56:56 +0000 Subject: Connectionists: [1st CfP] AAMAS 2027 Main track Message-ID: <7D502902-CBA5-4E0C-A2B3-20EA70252BC8@uu.nl> Dear all, We are happy to announce that the submission for AAMAS 2027 is now open! Key dates: Author registration on OpenReview: 17 September 2026 Abstract submission: 1 October 2026 Paper submission: 8 October 2026 Conference website: https://warwick.ac.uk/fac/sci/dcs/aamas2027/ Call for Papers: https://warwick.ac.uk/fac/sci/dcs/aamas2027/calls/ OpenReview: https://openreview.net/group?id=ifaamas.org/AAMAS/2027/Conference The 26th International Conference on Autonomous Agents and Multiagent Systems will take place in Hanoi, Vietnam, from 3 to 7 May 2027. AAMAS welcomes research across the full breadth of autonomous agents and multiagent systems, covering established areas as well as emerging topics such as generative and agentic AI. A new feature for 2027: *Findings of AAMAS* An important novelty this year is the introduction of the Findings of AAMAS 2027, following a model successfully adopted by the ACL community. Our motivation is simple. The agents and multiagent systems community produces a large number and a remarkable variety of valuable research contributions. A single, highly selective proceedings volume cannot always accommodate all the work that is technically sound, scientifically useful, and capable of stimulating further research. We would like more of this work to become visible, citable, and available for discussion within the community. We hope that the Findings will help valuable contributions reach AAMAS participants, generate feedback, inspire new ideas, and encourage follow-up research and collaborations. In practical terms, AAMAS 2027 will have two archival publication volumes: - the Proceedings of AAMAS 2027, containing papers selected for the main conference proceedings; - the Findings of AAMAS 2027, providing an additional publication venue for papers not selected for the Proceedings but nevertheless judged worthy of publication. Authors will not need to submit separately to the Findings. Papers not selected for the Proceedings will automatically be considered for the Findings, unless the authors opt out during submission. Papers selected for the Proceedings and papers selected for the Findings will have the same length and follow the same submission format. Both volumes will be published under a CC BY licence. For more information: https://warwick.ac.uk/fac/sci/dcs/aamas2027/calls/findings/ *More space for discussion* Papers accepted for the Proceedings will be presented both in an oral session and during a poster session. Papers accepted for the Findings will be presented during a poster session. Building on AAMAS?s long-standing tradition of lively scientific exchange, we intend to give ample space to discussion, interaction, and networking throughout the conference. In particular, we want poster presentations to be a central part of the scientific programme, giving participants meaningful opportunities to engage with authors, discuss results and open questions, exchange feedback, discover connections across different research areas, and develop new ideas and collaborations. All submissions will undergo rigorous peer review and will be evaluated on criteria including originality, significance, soundness, reproducibility, clarity, relevance to the conference, presentation quality, and appropriate engagement with the state of the art. Important dates: Author registration on OpenReview: 17 September 2026 Abstract submission: 1 October 2026 Paper submission: 8 October 2026 Rebuttal period: 20?24 November 2026 Author notification: 21 December 2026 Camera-ready deadline: 25 January 2027 Conference: 3?7 May 2027 All submission deadlines are Anywhere on Earth, UTC?12. Please have a look at the full Call for Papers, start preparing your submissions, and keep an eye on the website for the opening of the OpenReview submission site. Please also feel free to forward this announcement to colleagues, research groups, and mailing lists that may be interested. We are excited about this new chapter for AAMAS and look forward to receiving your submissions, and to discussing your work with you in Hanoi! Best wishes, AAMAS 2027 Organising Team -------------- next part -------------- An HTML attachment was scrubbed... URL: From ASIM.ROY at asu.edu Fri Aug 21 00:16:37 2026 From: ASIM.ROY at asu.edu (Asim Roy) Date: Fri, 21 Aug 2026 04:16:37 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Charlie Gross was in that discussion with Horace Barlow. So was Itzhak Fried and Christof Koch. The referenced Jennifer Aniston experiments were conducted under their guidance at UCLA. And so was Walter Freeman and many other prominent neuro and cognitive scientists involved in that discussion. Gross (2002) characterized grandmother cells as ?multimodal.? Reddy and Thorpe (2014) also define concepts cells as ?invariant, multimodal.? Thus, the fundamental characterization of these two concepts is identical. ?Multimodal invariance? implies associative learning across modalities. They also state that concept cells have ?meaning of a given stimulus in a manner that is invariant to different representations of that stimulus.? ?Selectivity or specificity,? ?complex concept,? ?meaning,? ?multimodal invariance? and ?abstractness? (Reddy and Thorpe, 2014) are all integral properties of both concept cells and grandmother cells. Reddy and Thorpe (2014) categorically state that ?abstract, invariant representations? is ?a hallmark of MTL concept cells.? Quiroga (2012) also admit that concept cells are abstract: ?These and many other examples suggest that MTL neurons encode an abstract representation of the concept triggered by the stimulus.? Unfortunately, Barwich (2019) is completely flawed in its arguments against grandmother cells as argued in Roy?s commentary on the article (Roy (2020)). The neurophysiological evidence for purely abstract single cell representation in the brain points to symbolic representation in the brain. Quiroga et al. (2008) estimate that 40% of MTL cells are abstract. Thus, there is extensive use of abstractions at the single cell level in the brain. This is the neurophysiological evidence for the neurosymbolic approach and what Gary Marcus has been arguing for all along. In a way, both approaches win and need each other. 1. Barlow H. B. (2009). Grandmother cells, symmetry, and invariance: how the term arose and what the facts suggest, in The Cognitive Neurosciences, 4th Edn., ed GazzanigaM. (Cambridge, MA: MIT Press), 309?320. 2. Gross C. G. (2002). Genealogy of the ?grandmother cell?. Neuroscientist8, 512?518. 10.1016/j.meegid.2016.04.006 3. Reddy L. Thorpe S. J. (2014). Concept cells through associative learning of high-level representations. Neuron84, 248?251. 10.1016/j.neuron.2014.10.004 4. Barwich A-S (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 5. Roy A (2020) Commentary: The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 14:59. doi: 10.3389/fnins.2020.00059 6. Quiroga R. Q. Kreiman G. Koch C.FriedI. (2008). Sparse but not ?grandmother-cell? coding in the medial temporal lobe. Trends Cogn. Sci.12, 87?91. 10.1016/j.tics.2007.12.003 Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: KENTRIDGE, ROBERT W. Sent: Thursday, August 20, 2026 3:53 AM To: Asim Roy ; Gary Marcus ; Stephen Jos? Hanson Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... It is worth pointing out that in Jerry Lettvin?s original paper on the grandmother cell the concept was presented as a joke. Lettvin certainly believed in the notion that one might identify cells in the frog?s brain as ?bug detectors?, but he appeared to be sceptical about the idea that such cells might exist in the human brain representing complex concepts like ?my grandmother?. There are a couple of great papers on the history of the grandmother cell: Ann-Sophie Barwich (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 Charles C Gross (2002) Genealogy of the "grandmother cell?. Neuroscientist 8(5):512-8. doi: 10.1177/107385802237175. I remember chatting to Charlie Gross about this around that time (he also reminisced about his part-time job mowing BF Skinner?s lawn). People often assume that the Quiroga et al (2005 Nature, 435. 1102 - 1107) paper on the Jennifer Aniston cell was positive evidence for the existence of grandmother cells in the human brain. This is far from the truth. The Jennifer Aniston cell does not only respond to Jennifer Aniston, but also to related concepts like ?Lisa Kudow? (another actress in ?Friends?), so activity in that cell does not uniquely code ?Jennifer Aniston? and cannot be used to read out the presence of Jennifer Aniston. Quiroga et al make this clear in the discussion of their findings. Bob Kentridge Psychology Department University of Durham Canadian Institute for Advanced Research Programme on Brain, Mind and Consciousness Sent from Outlook for Mac From: Connectionists > on behalf of Asim Roy > Date: Thursday, 20 August 2026 at 06:51 To: Gary Marcus >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] The Horace Barlow note was after a furious private debate with some eminent neuro and cognitive scientists regarding Jennifer Aniston cells being evidence for grandmother cells. And this was at age 93 when he visited me after defending his theory. At that stage of his career, his insights were profound. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Asim Roy Sent: Wednesday, August 19, 2026 6:28 PM To: 'Gary Marcus' >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: FW: AI at 50 videos... I had some communication with Marvin Minsky and Jerry Fodor years ago. Here are two quotes from them: 1. Marvin Minsky: "Don?t pay any attention to the critics. Don?t even ignore them.? -By Sam Goldwyn 2. Jerry Fodor: ?Arguing with Connectionists is like arguing with zombies; both are dead, but neither has noticed it.? And here?s a note from a neuroscience pioneer Horace Barlow, Darwin?s great-grandson (Horace Barlow - Wikipedia), winner of many neuroscience prizes. Without his pioneering work poking into frog brains with microelectrodes, we would NOT have AI as we see it today. The grandmother cell theory essentially points to where cognition exists in an abstract form. And there is plenty of neurophysiological evidence for such single cell abstractions, starting with line orientation cells. ====================================================================================================================================== Dear Asim Yes, I would like to join your group, though I would (again) like to make it clear that the grandmother cell theory is not mine, though I still hold that something conforming amazingly well to what was conceptualised by Jerry Lettvin 50 years ago, really does exist! I am still actively interested in the topic, and have just sent off the final proofs of an essay bearing on it. I shall send you a copy of the final proof as soon as I am allowed to circulate it, and of course the more widely discussed it is the better pleased I shall be, though I fear that what I have written will not be universally accepted, at least at first! Best regards Horace ============================================================================================================= Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Connectionists > On Behalf Of Gary Marcus Sent: Wednesday, August 19, 2026 11:23 AM To: Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... yep: neural networks ascended, pushed symbols out of the way, and then symbols came back relabeled as harnesses, loops code interpreters, and tools. neurosymbolic AI, the hybrid of the two approaches, clearly beat either of the two alternatives on their own. claude code, biggest recent advance, shows that pretty clearly, when you look under the hood. gary On Aug 19, 2026, at 10:16?AM, Stephen Jos? Hanson > wrote: ? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- -------------- next part -------------- An HTML attachment was scrubbed... URL: From samuel.kaski at manchester.ac.uk Fri Aug 21 07:01:19 2026 From: samuel.kaski at manchester.ac.uk (Samuel Kaski) Date: Fri, 21 Aug 2026 11:01:19 +0000 Subject: Connectionists: ELLIS Institute Finland hiring: postdocs and PhD students in AI and machine learning. DL Sept. 21 Message-ID: ELLIS Institute Finland is hiring postdocs and PhD students in AI and machine learning. DL Sept. 21. - Cutting-edge computational resources like LUMI AI Factory - Close collab with ELLIS - Ambitious, high-impact projects with ~50 institute faculty https://www.ellisinstitute.fi/postdoc-and-phd-recruit-autumn-2026 Want to turn breakthrough research into a competitive advantage for a startup? ELLIS Institute Finland offers an entrepreneurial postdoc track for recent PhDs who want to evolve science into real-world ventures, for instance in AI4Science. Apply by Sept. 21: https://www.ellisinstitute.fi/entrepreneurial-postdoc-recruit-autumn-2026 -------------- next part -------------- An HTML attachment was scrubbed... URL: From marysue0301 at gmail.com Fri Aug 21 04:13:52 2026 From: marysue0301 at gmail.com (Ruichen CONG) Date: Fri, 21 Aug 2026 16:13:52 +0800 Subject: Connectionists: CFP (Late Breaking Submission Open to Aug. 25, 2026): The 11th IEEE Cyber Science and Technology Congress (CyberSciTech 2026) - Melbourne, Australia, November 9-13, 2026 Message-ID: [Our apologies if you receive multiple copies of this CFP] Dear Colleagues, We are writing to invite you to submit your papers at *The 11th IEEE Cyber Science and Technology Congress (CyberSciTech 2026)scheduled on November 9-13, 2026, Melbourne, Australia.* https://cyber-science.org/2026/cyberscitech/ *The Late Breaking Submission Deadline is quickly approaching on August 25, 2026.* We look forward to your submissions! Submission link: https://edas.info/N35527 IMPORTANT DATES ----------------------------------------------- Regular Paper Submission Due: Aug. 15, 2026 (Extended) *WiP/Poster/Wksp/SS Paper Due: Aug. 25, 2026 (Extended)Late Breaking Submission Due: Aug. 25, 2026 (Extended)* Author Notification: Sep. 01, 2026 Paper Registration Due: Sep. 24, 2026 Camera-ready Submission Due: Oct. 01, 2026 CONGRESS INTRODUCTION -------------------------- Cyberspace, the seamless integration of physical, social, and mental spaces, is an integral part of our society, ranging from learning and entertainment to business and cultural activities, and so on. There are, however, in addition to its technical challenges, a number of pressing issues such as safety and trust associated with the cyberspace. To address these challenges, there is a need to establish new science and research portfolios that incorporate cyber-physical, cyber-social, cyber-intelligent, and cyber-life technologies in a cohesive and efficient manner. This is the aim of the IEEE Cyber Science and Technology Congress (CyberSciTech). IEEE CyberSciTech has been successfully held in Auckland, New Zealand, 2016, in Orlando, USA, 2017, in Athens, Greece, 2018, in Fukuoka, Japan, 2019, in Calgary, Canada, 2020 and 2021 (online due to COVID-19), in Calabria, Italy, 2022, in Abu Dhabi, UAE, 2023, in Boracay Island, Philippines, 2024, and in Hakodate City, Japan, 2025. In 2026, we will continue to offer IEEE CyberSciTech with the aim of providing a common platform for scientists, researchers, and engineers to share their latest ideas and advances in the broad scope of cyber-related science, technology, and application topics. In addition, this is also a platform to allow relevant stakeholders to get together, discuss and identify ongoing and emerging challenges, in order to understand and shape new cyber-enabled worlds. SCOPE AND TRACKSs ----------------------------------------------- Topics of interest include, but are not limited to: Regular Tracks (6-8 pages) Track 1: Cyberspace Theory & Technology - Cyberspace Property, Structure & Models - Cyber Pattern, Evolution, Ecology & Science - SDN/SDS, 5G/6G, Vehicle & Novel Network - Cloud, Fog, Edge & Green Computing - Big Data Analytics, Technology & Service - Infrastructures for Smart City/Country Track 2: Cyber Security, Privacy & Trust - Cyber Security, Safety & Resilience - Cyber Crime, Fraud, Abuse & Forensics - Cyber Attack, Terrorism, Warfare & Defense - Cyber Privacy, Trust & Insurance - Blockchain, DLT Techniques & Applications - Post-quantum Cryptography Track 3: Cyber Physical Computing & Systems - Cyber Physical Systems & Interfaces - Cyber Physical Dynamics & Disaster Relief - Cyber Manufacturing & Control - Embedded Systems & Software - Autonomous Robots & Vehicles - IoT, Digital Twin & Smart Systems Track 4: Cyber Social Computing & Networks - Social Networking & Computing - Computational Social Science - Crowd Sourcing, Sensing & Computing - Cyber Culture, Relation, Creation & Art - Cyber Social Right, Policy, Laws & Ethics - Cyber Learning, Economics & Politics Track 5: Cyber Intelligence & Cognitive Science - Cyber/Digital Brain & Artificial Intelligence - Hybrid & Hyper-connected Intelligence - Affective/Mind Cognition & Computing - Brain/Mind Machine Interface - AI Agents & Embodied Intelligence - Intelligent Object, Environment & Service Track 6: Cyber Life & Wellbeing - Cyber Life & Human Centric Computing - Cyber Medicine, Healthcare & Psychology - Cyborg/Wearable/Implantable Technology - Human/Animal Behavior Recognition - Personal Big Data & Personality Computing - Augmented/Mixed Reality & Metaverse IEEE CyberSciTech 2026 CALLS ---------------------------------- For original papers in: - Regular Tracks: 6-8 pages - WiP/Workshop/Special Session Tracks: 4-6 pages - Poster Track: 2 pages - LBI (Late Breaking Innovation) Track: 4-8 pages - All accepted conference, workshop, special session (SS), and poster papers will be published by IEEE in the Conference Proceedings (IEEE-DL and EI indexed). Selected high quality papers will be recommended to prestige journal special issues. Submission link: https://edas.info/N35527 SUBMISSION GUIDELINES ---------------------------------- Authors are invited to submit their original work that has not previously been submitted or published in any other venue. Regular, Work-in-Progress (WiP), Workshop/SS, Poster papers all need to be in IEEE CS format ( https://www.ieee.org/conferences/publishing/templates.html) CO-LOCATED CONFERENCES ---------------------------------- - The 24th IEEE International Conference on Pervasive Intelligence and Computing (PICom 2026) - The 24th IEEE International Conference on Dependable, Autonomic and Secure Computing (DASC 2026) - The 12th IEEE International Conference on Cloud and Big Data Computing (CBDCom 2026) Hosted by RMIT University, Australia -------------- next part -------------- An HTML attachment was scrubbed... URL: From jon at scios.tech Fri Aug 21 05:43:17 2026 From: jon at scios.tech (Jonathan Starr) Date: Fri, 21 Aug 2026 05:43:17 -0400 Subject: Connectionists: IOSP 2026 in Leiden: build a community-owned open science system Message-ID: The Institute of Open Science Practices is hosting its 2026 workshops in Leiden from October 12 to 15, and we'd like you in the room. *It's free to attend.* Across the two main working days, the event will establish a participant-owned data storage consortium, publish datasets, code, and knowledge objects to it, create our own algorithms that assess the trust and value of what's published, and assemble community-owned collaboratives that route funding based on their interpretation of that analysis. The result will be a fully functional open science system, owned by the community that builds it, to test, build on, and iterate over the coming months and years. The pieces of open science already exist. It's time they were woven into a coherent system. You can already see the emerging data network here: https://www.iosp.science/datanetwork And the full workshop line-up is here: https://iosp.science/workshops Four ways to participate: 1. *Register.* The room holds 100 people, and last year 425 registered for an 80-person room, so a registration isn't yet a seat and we confirm by email. The form asks whether you'd need travel support; say so if you would. https://www.iosp.science/?signup=participant 2. *Contribute to the community track.* It runs alongside the main track on both working days and is an open space for talks, workshops, panels, and discussions around the four themes. https://www.iosp.science/submit-community-session 3. *Building a tool? Submit it to the showcase.* We'll stress-test it and build on it in Leiden. https://www.iosp.science/?signup=showcase 4. *Join the resilient data workshop.* The workshop that opens IOSP 2026 has you stand up an IPFS node and join a member-owned resilient data consortium. Your node will run either on your own laptop or on a Raspberry Pi kit you take home. There are only 20 seats and 10 take-home Raspberry Pi kits. Earlier sign-ups get first consideration if it fills. Signing up for this workshop also registers you for IOSP 2026. https://www.iosp.science/resilient-data-signup And finally, spread the word. Forward this, post the community-track call, or point a builder at the workshop or showcase. *Hope to see you in Leiden!* Jon, Ellie, and the IOSP organizing team *P.S.* The resilient data consortium needs data worth keeping. If you know a public dataset that could vanish?a shuttered project's archive, a retiring server, data whose funding ended?recommend it and we'll consider adding it to the network. Recommend a dataset: https://www.iosp.science/resilient-data-signup See what the network holds now: https://www.iosp.science/datanetwork -------------- next part -------------- An HTML attachment was scrubbed... URL: From ASIM.ROY at asu.edu Fri Aug 21 00:20:02 2026 From: ASIM.ROY at asu.edu (Asim Roy) Date: Fri, 21 Aug 2026 04:20:02 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Hava, you are not dead. Those statements of Marvin Minsky and Jerry Fodor reflect the bitterness of those days. I have also heard a connectionist call Marvin a ?devil? in an acceptance speech for an award. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Hava Siegelmann Sent: Thursday, August 20, 2026 8:29 AM To: Asim Roy Cc: Gary Marcus ; Stephen Jos? Hanson ; director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... This is amazing I love it - dead ! On Thu, Aug 20, 2026 at 8:30?AM Asim Roy > wrote: I had some communication with Marvin Minsky and Jerry Fodor years ago. Here are two quotes from them: 1. Marvin Minsky: "Don?t pay any attention to the critics. Don?t even ignore them.? -By Sam Goldwyn 2. Jerry Fodor: ?Arguing with Connectionists is like arguing with zombies; both are dead, but neither has noticed it.? And here?s a note from a neuroscience pioneer Horace Barlow, Darwin?s great-grandson (Horace Barlow - Wikipedia), winner of many neuroscience prizes. Without his pioneering work poking into frog brains with microelectrodes, we would NOT have AI as we see it today. The grandmother cell theory essentially points to where cognition exists in an abstract form. And there is plenty of neurophysiological evidence for such single cell abstractions, starting with line orientation cells. ====================================================================================================================================== Dear Asim Yes, I would like to join your group, though I would (again) like to make it clear that the grandmother cell theory is not mine, though I still hold that something conforming amazingly well to what was conceptualised by Jerry Lettvin 50 years ago, really does exist! I am still actively interested in the topic, and have just sent off the final proofs of an essay bearing on it. I shall send you a copy of the final proof as soon as I am allowed to circulate it, and of course the more widely discussed it is the better pleased I shall be, though I fear that what I have written will not be universally accepted, at least at first! Best regards Horace ============================================================================================================= Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Connectionists > On Behalf Of Gary Marcus Sent: Wednesday, August 19, 2026 11:23 AM To: Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... yep: neural networks ascended, pushed symbols out of the way, and then symbols came back relabeled as harnesses, loops code interpreters, and tools. neurosymbolic AI, the hybrid of the two approaches, clearly beat either of the two alternatives on their own. claude code, biggest recent advance, shows that pretty clearly, when you look under the hood. gary On Aug 19, 2026, at 10:16?AM, Stephen Jos? Hanson > wrote: ? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- -- -- Hava T. Siegelmann, Ph.D. Provost Professor Director, BINDS Lab (Biologically Inspired Neural Dynamical Systems) Dept. of Computer Science Core member, Program of Neuroscience and Behavior University of Massachusetts Amherst Amherst, MA, 01003 Phone: 413-540-6826 LAB WEBSITE: http://binds.cs.umass.edu/ -------------- next part -------------- An HTML attachment was scrubbed... URL: From robert.kentridge at durham.ac.uk Fri Aug 21 12:14:27 2026 From: robert.kentridge at durham.ac.uk (KENTRIDGE, ROBERT W.) Date: Fri, 21 Aug 2026 16:14:27 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: I don?t think there is much I disagree with in Asim?s comment. I does go to show, though, how important the details of one?s definition of a grandmother cell is. If you take the hardline definition that a grandmother cell must respond to one concept and one concept alone, so that you can read off the presence of an ?x? from the response of a grandmother cell tuned to ?x?s then the Jennifer Aniston cell is not a grandmother cell. If you take Asim?s definition that they are cells which respond to a complex concept across a range of modalities etc., without any proviso that they must respond exclusively to ?x?s then the Jennifer Aniston cell is indeed a grandmother cell. Reddy and Thorpe certainly describe an interesting class of cell, I?m just not sure if they correspond to Lettvin?s view. BTW, I talked to Charlie Gross about all this in the 1980s or 90s - I haven?t talked to him about this more recently so his views might have changed. I also knew Horace Barlow a bit, but my conversations with him were about artificial neural nets and computation (again in the 80s and 90s), not neurophysiology. At the first talk I ever gave I notice Horace in the audience and shuddered - he was a bit terrifying to a young neophyte. He did ask a question at the end of my talk, but it wasn?t one that stumped me, and he was very gracious about the whole thing. cheers, Bob Sent from Outlook for Mac From: Asim Roy Date: Friday, 21 August 2026 at 05:16 To: KENTRIDGE, ROBERT W. ; Gary Marcus ; Stephen Jos? Hanson ; Hava Siegelmann (hava.siegelmann at gmail.com) ; Juergen Schmidhuber ; Ali Minai Cc: director at inc.ucsd.edu ; connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] Charlie Gross was in that discussion with Horace Barlow. So was Itzhak Fried and Christof Koch. The referenced Jennifer Aniston experiments were conducted under their guidance at UCLA. And so was Walter Freeman and many other prominent neuro and cognitive scientists involved in that discussion. Gross (2002) characterized grandmother cells as ?multimodal.? Reddy and Thorpe (2014) also define concepts cells as ?invariant, multimodal.? Thus, the fundamental characterization of these two concepts is identical. ?Multimodal invariance? implies associative learning across modalities. They also state that concept cells have ?meaning of a given stimulus in a manner that is invariant to different representations of that stimulus.? ?Selectivity or specificity,? ?complex concept,? ?meaning,? ?multimodal invariance? and ?abstractness? (Reddy and Thorpe, 2014) are all integral properties of both concept cells and grandmother cells. Reddy and Thorpe (2014) categorically state that ?abstract, invariant representations? is ?a hallmark of MTL concept cells.? Quiroga (2012) also admit that concept cells are abstract: ?These and many other examples suggest that MTL neurons encode an abstract representation of the concept triggered by the stimulus.? Unfortunately, Barwich (2019) is completely flawed in its arguments against grandmother cells as argued in Roy?s commentary on the article (Roy (2020)). The neurophysiological evidence for purely abstract single cell representation in the brain points to symbolic representation in the brain. Quiroga et al. (2008) estimate that 40% of MTL cells are abstract. Thus, there is extensive use of abstractions at the single cell level in the brain. This is the neurophysiological evidence for the neurosymbolic approach and what Gary Marcus has been arguing for all along. In a way, both approaches win and need each other. 1. Barlow H. B. (2009). Grandmother cells, symmetry, and invariance: how the term arose and what the facts suggest, in The Cognitive Neurosciences, 4th Edn., ed GazzanigaM. (Cambridge, MA: MIT Press), 309?320. 2. Gross C. G. (2002). Genealogy of the ?grandmother cell?. Neuroscientist8, 512?518. 10.1016/j.meegid.2016.04.006 3. Reddy L. Thorpe S. J. (2014). Concept cells through associative learning of high-level representations. Neuron84, 248?251. 10.1016/j.neuron.2014.10.004 4. Barwich A-S (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 5. Roy A (2020) Commentary: The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 14:59. doi: 10.3389/fnins.2020.00059 6. Quiroga R. Q. Kreiman G. Koch C.FriedI. (2008). Sparse but not ?grandmother-cell? coding in the medial temporal lobe. Trends Cogn. Sci.12, 87?91. 10.1016/j.tics.2007.12.003 Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: KENTRIDGE, ROBERT W. Sent: Thursday, August 20, 2026 3:53 AM To: Asim Roy ; Gary Marcus ; Stephen Jos? Hanson Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... It is worth pointing out that in Jerry Lettvin?s original paper on the grandmother cell the concept was presented as a joke. Lettvin certainly believed in the notion that one might identify cells in the frog?s brain as ?bug detectors?, but he appeared to be sceptical about the idea that such cells might exist in the human brain representing complex concepts like ?my grandmother?. There are a couple of great papers on the history of the grandmother cell: Ann-Sophie Barwich (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 Charles C Gross (2002) Genealogy of the "grandmother cell?. Neuroscientist 8(5):512-8. doi: 10.1177/107385802237175. I remember chatting to Charlie Gross about this around that time (he also reminisced about his part-time job mowing BF Skinner?s lawn). People often assume that the Quiroga et al (2005 Nature, 435. 1102 - 1107) paper on the Jennifer Aniston cell was positive evidence for the existence of grandmother cells in the human brain. This is far from the truth. The Jennifer Aniston cell does not only respond to Jennifer Aniston, but also to related concepts like ?Lisa Kudow? (another actress in ?Friends?), so activity in that cell does not uniquely code ?Jennifer Aniston? and cannot be used to read out the presence of Jennifer Aniston. Quiroga et al make this clear in the discussion of their findings. Bob Kentridge Psychology Department University of Durham Canadian Institute for Advanced Research Programme on Brain, Mind and Consciousness Sent from Outlook for Mac From: Connectionists > on behalf of Asim Roy > Date: Thursday, 20 August 2026 at 06:51 To: Gary Marcus >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] The Horace Barlow note was after a furious private debate with some eminent neuro and cognitive scientists regarding Jennifer Aniston cells being evidence for grandmother cells. And this was at age 93 when he visited me after defending his theory. At that stage of his career, his insights were profound. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Asim Roy Sent: Wednesday, August 19, 2026 6:28 PM To: 'Gary Marcus' >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: FW: AI at 50 videos... I had some communication with Marvin Minsky and Jerry Fodor years ago. Here are two quotes from them: 1. Marvin Minsky: "Don?t pay any attention to the critics. Don?t even ignore them.? -By Sam Goldwyn 2. Jerry Fodor: ?Arguing with Connectionists is like arguing with zombies; both are dead, but neither has noticed it.? And here?s a note from a neuroscience pioneer Horace Barlow, Darwin?s great-grandson (Horace Barlow - Wikipedia), winner of many neuroscience prizes. Without his pioneering work poking into frog brains with microelectrodes, we would NOT have AI as we see it today. The grandmother cell theory essentially points to where cognition exists in an abstract form. And there is plenty of neurophysiological evidence for such single cell abstractions, starting with line orientation cells. ====================================================================================================================================== Dear Asim Yes, I would like to join your group, though I would (again) like to make it clear that the grandmother cell theory is not mine, though I still hold that something conforming amazingly well to what was conceptualised by Jerry Lettvin 50 years ago, really does exist! I am still actively interested in the topic, and have just sent off the final proofs of an essay bearing on it. I shall send you a copy of the final proof as soon as I am allowed to circulate it, and of course the more widely discussed it is the better pleased I shall be, though I fear that what I have written will not be universally accepted, at least at first! Best regards Horace ============================================================================================================= Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Connectionists > On Behalf Of Gary Marcus Sent: Wednesday, August 19, 2026 11:23 AM To: Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... yep: neural networks ascended, pushed symbols out of the way, and then symbols came back relabeled as harnesses, loops code interpreters, and tools. neurosymbolic AI, the hybrid of the two approaches, clearly beat either of the two alternatives on their own. claude code, biggest recent advance, shows that pretty clearly, when you look under the hood. gary On Aug 19, 2026, at 10:16?AM, Stephen Jos? Hanson > wrote: ? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- -------------- next part -------------- An HTML attachment was scrubbed... URL: From jose at rubic.rutgers.edu Fri Aug 21 13:52:33 2026 From: jose at rubic.rutgers.edu (=?utf-8?B?U3RlcGhlbiBKb3PDqSBIYW5zb24=?=) Date: Fri, 21 Aug 2026 17:52:33 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: <75a7a52b-83c7-4e3a-a38d-1167051f1ba3@rubic.rutgers.edu> Yup some bright lights are going out way too often now. I meant Larry at Oxford when I was on the McDonnell-PEW CNS advisory board. Very kind fellow. Charlie was also a great Historian and loved to put things in their proper context. Best Stephen On 8/21/26 13:46, KENTRIDGE, ROBERT W. wrote: Hi Stephen, Your recollections and mine of Charlie?s views seem very similar - Charlie did not believe his face cells and the like were grandmother cells - they were part of a population code rather than being cells whose activity could be read out as unambiguous evidence for the presence of ?x?s. So, his concept must have been closer to my ?hardline? definition than to Reddy and Thorpe?s conception. I knew that Charlie had died in 2019. I think the last time I saw him was at a meeting in Oxford, I don?t remember the date, with Alan Cowey (alas another great, and great friend, we have lost, along with Larry Weiskrantz who was one of my closest friends in visual neuroscience - I published a lot with both of them). At that meeting Charlie was clearly far from well - so sad. cheers, Bob Sent from Outlook for Mac From: Stephen Jos? Hanson Date: Friday, 21 August 2026 at 17:28 To: KENTRIDGE, ROBERT W. ; Asim Roy ; Gary Marcus ; Hava Siegelmann (hava.siegelmann at gmail.com) ; Juergen Schmidhuber ; Ali Minai Cc: director at inc.ucsd.edu ; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] Hi Robert, I was a friend of Charlies in princeton from the 80s. Charlie died in 2019. And we had many dinners together.. occasionally with his most recent wife, Joyce Carol Oats. Charlie argued his face cells (and hand cells0 were not grandmother cells: they were coarsely tuned, responded to many faces, and fit an ensemble/population code. Its hard to fight a counterfactual, as there are too many ways a concept or category intension could be false. How many control conditions do you need before it is categorical response to a specific stimulus? Is an equivalence class enough for it to be categorical? Same problem with so called fuisform face area.. A relevant paper to this is something I wrote a few years ago in Frontiers, Human Neuroscience.. The The Failure of Blobology: fMRI Misinterpretation, Maleficience and Muddle https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2022.870091/full On 8/21/26 12:14, KENTRIDGE, ROBERT W. wrote: I don?t think there is much I disagree with in Asim?s comment. I does go to show, though, how important the details of one?s definition of a grandmother cell is. If you take the hardline definition that a grandmother cell must respond to one concept and one concept alone, so that you can read off the presence of an ?x? from the response of a grandmother cell tuned to ?x?s then the Jennifer Aniston cell is not a grandmother cell. If you take Asim?s definition that they are cells which respond to a complex concept across a range of modalities etc., without any proviso that they must respond exclusively to ?x?s then the Jennifer Aniston cell is indeed a grandmother cell. Reddy and Thorpe certainly describe an interesting class of cell, I?m just not sure if they correspond to Lettvin?s view. BTW, I talked to Charlie Gross about all this in the 1980s or 90s - I haven?t talked to him about this more recently so his views might have changed. I also knew Horace Barlow a bit, but my conversations with him were about artificial neural nets and computation (again in the 80s and 90s), not neurophysiology. At the first talk I ever gave I notice Horace in the audience and shuddered - he was a bit terrifying to a young neophyte. He did ask a question at the end of my talk, but it wasn?t one that stumped me, and he was very gracious about the whole thing. cheers, Bob Sent from Outlook for Mac From: Asim Roy Date: Friday, 21 August 2026 at 05:16 To: KENTRIDGE, ROBERT W. ; Gary Marcus ; Stephen Jos? Hanson ; Hava Siegelmann (hava.siegelmann at gmail.com) ; Juergen Schmidhuber ; Ali Minai Cc: director at inc.ucsd.edu ; connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] Charlie Gross was in that discussion with Horace Barlow. So was Itzhak Fried and Christof Koch. The referenced Jennifer Aniston experiments were conducted under their guidance at UCLA. And so was Walter Freeman and many other prominent neuro and cognitive scientists involved in that discussion. Gross (2002) characterized grandmother cells as ?multimodal.? Reddy and Thorpe (2014) also define concepts cells as ?invariant, multimodal.? Thus, the fundamental characterization of these two concepts is identical. ?Multimodal invariance? implies associative learning across modalities. They also state that concept cells have ?meaning of a given stimulus in a manner that is invariant to different representations of that stimulus.? ?Selectivity or specificity,? ?complex concept,? ?meaning,? ?multimodal invariance? and ?abstractness? (Reddy and Thorpe, 2014) are all integral properties of both concept cells and grandmother cells. Reddy and Thorpe (2014) categorically state that ?abstract, invariant representations? is ?a hallmark of MTL concept cells.? Quiroga (2012) also admit that concept cells are abstract: ?These and many other examples suggest that MTL neurons encode an abstract representation of the concept triggered by the stimulus.? Unfortunately, Barwich (2019) is completely flawed in its arguments against grandmother cells as argued in Roy?s commentary on the article (Roy (2020)). The neurophysiological evidence for purely abstract single cell representation in the brain points to symbolic representation in the brain. Quiroga et al. (2008) estimate that 40% of MTL cells are abstract. Thus, there is extensive use of abstractions at the single cell level in the brain. This is the neurophysiological evidence for the neurosymbolic approach and what Gary Marcus has been arguing for all along. In a way, both approaches win and need each other. 1. Barlow H. B. (2009). Grandmother cells, symmetry, and invariance: how the term arose and what the facts suggest, in The Cognitive Neurosciences, 4th Edn., ed GazzanigaM. (Cambridge, MA: MIT Press), 309?320. 2. Gross C. G. (2002). Genealogy of the ?grandmother cell?. Neuroscientist8, 512?518. 10.1016/j.meegid.2016.04.006 3. Reddy L. Thorpe S. J. (2014). Concept cells through associative learning of high-level representations. Neuron84, 248?251. 10.1016/j.neuron.2014.10.004 4. Barwich A-S (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 5. Roy A (2020) Commentary: The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 14:59. doi: 10.3389/fnins.2020.00059 6. Quiroga R. Q. Kreiman G. Koch C.FriedI. (2008). Sparse but not ?grandmother-cell? coding in the medial temporal lobe. Trends Cogn. Sci.12, 87?91. 10.1016/j.tics.2007.12.003 Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: KENTRIDGE, ROBERT W. Sent: Thursday, August 20, 2026 3:53 AM To: Asim Roy ; Gary Marcus ; Stephen Jos? Hanson Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... It is worth pointing out that in Jerry Lettvin?s original paper on the grandmother cell the concept was presented as a joke. Lettvin certainly believed in the notion that one might identify cells in the frog?s brain as ?bug detectors?, but he appeared to be sceptical about the idea that such cells might exist in the human brain representing complex concepts like ?my grandmother?. There are a couple of great papers on the history of the grandmother cell: Ann-Sophie Barwich (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 Charles C Gross (2002) Genealogy of the "grandmother cell?. Neuroscientist 8(5):512-8. doi: 10.1177/107385802237175. I remember chatting to Charlie Gross about this around that time (he also reminisced about his part-time job mowing BF Skinner?s lawn). People often assume that the Quiroga et al (2005 Nature, 435. 1102 - 1107) paper on the Jennifer Aniston cell was positive evidence for the existence of grandmother cells in the human brain. This is far from the truth. The Jennifer Aniston cell does not only respond to Jennifer Aniston, but also to related concepts like ?Lisa Kudow? (another actress in ?Friends?), so activity in that cell does not uniquely code ?Jennifer Aniston? and cannot be used to read out the presence of Jennifer Aniston. Quiroga et al make this clear in the discussion of their findings. Bob Kentridge Psychology Department University of Durham Canadian Institute for Advanced Research Programme on Brain, Mind and Consciousness Sent from Outlook for Mac From: Connectionists > on behalf of Asim Roy > Date: Thursday, 20 August 2026 at 06:51 To: Gary Marcus >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] The Horace Barlow note was after a furious private debate with some eminent neuro and cognitive scientists regarding Jennifer Aniston cells being evidence for grandmother cells. And this was at age 93 when he visited me after defending his theory. At that stage of his career, his insights were profound. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Asim Roy Sent: Wednesday, August 19, 2026 6:28 PM To: 'Gary Marcus' >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: FW: AI at 50 videos... I had some communication with Marvin Minsky and Jerry Fodor years ago. Here are two quotes from them: 1. Marvin Minsky: "Don?t pay any attention to the critics. Don?t even ignore them.? -By Sam Goldwyn 2. Jerry Fodor: ?Arguing with Connectionists is like arguing with zombies; both are dead, but neither has noticed it.? And here?s a note from a neuroscience pioneer Horace Barlow, Darwin?s great-grandson (Horace Barlow - Wikipedia), winner of many neuroscience prizes. Without his pioneering work poking into frog brains with microelectrodes, we would NOT have AI as we see it today. The grandmother cell theory essentially points to where cognition exists in an abstract form. And there is plenty of neurophysiological evidence for such single cell abstractions, starting with line orientation cells. ====================================================================================================================================== Dear Asim Yes, I would like to join your group, though I would (again) like to make it clear that the grandmother cell theory is not mine, though I still hold that something conforming amazingly well to what was conceptualised by Jerry Lettvin 50 years ago, really does exist! I am still actively interested in the topic, and have just sent off the final proofs of an essay bearing on it. I shall send you a copy of the final proof as soon as I am allowed to circulate it, and of course the more widely discussed it is the better pleased I shall be, though I fear that what I have written will not be universally accepted, at least at first! Best regards Horace ============================================================================================================= Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Connectionists > On Behalf Of Gary Marcus Sent: Wednesday, August 19, 2026 11:23 AM To: Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... yep: neural networks ascended, pushed symbols out of the way, and then symbols came back relabeled as harnesses, loops code interpreters, and tools. neurosymbolic AI, the hybrid of the two approaches, clearly beat either of the two alternatives on their own. claude code, biggest recent advance, shows that pretty clearly, when you look under the hood. gary On Aug 19, 2026, at 10:16?AM, Stephen Jos? Hanson > wrote: ? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- -- [cid:part1.5qPi40hB.1XKtzdcV at rubic.rutgers.edu] -- [cid:part1.5qPi40hB.1XKtzdcV at rubic.rutgers.edu] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.jpg Type: image/jpeg Size: 57383 bytes Desc: signature.jpg URL: From jose at rubic.rutgers.edu Fri Aug 21 12:28:16 2026 From: jose at rubic.rutgers.edu (=?utf-8?B?U3RlcGhlbiBKb3PDqSBIYW5zb24=?=) Date: Fri, 21 Aug 2026 16:28:16 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Hi Robert, I was a friend of Charlies in princeton from the 80s. Charlie died in 2019. And we had many dinners together.. occasionally with his most recent wife, Joyce Carol Oats. Charlie argued his face cells (and hand cells0 were not grandmother cells: they were coarsely tuned, responded to many faces, and fit an ensemble/population code. Its hard to fight a counterfactual, as there are too many ways a concept or category intension could be false. How many control conditions do you need before it is categorical response to a specific stimulus? Is an equivalence class enough for it to be categorical? Same problem with so called fuisform face area.. A relevant paper to this is something I wrote a few years ago in Frontiers, Human Neuroscience.. The The Failure of Blobology: fMRI Misinterpretation, Maleficience and Muddle https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2022.870091/full On 8/21/26 12:14, KENTRIDGE, ROBERT W. wrote: I don?t think there is much I disagree with in Asim?s comment. I does go to show, though, how important the details of one?s definition of a grandmother cell is. If you take the hardline definition that a grandmother cell must respond to one concept and one concept alone, so that you can read off the presence of an ?x? from the response of a grandmother cell tuned to ?x?s then the Jennifer Aniston cell is not a grandmother cell. If you take Asim?s definition that they are cells which respond to a complex concept across a range of modalities etc., without any proviso that they must respond exclusively to ?x?s then the Jennifer Aniston cell is indeed a grandmother cell. Reddy and Thorpe certainly describe an interesting class of cell, I?m just not sure if they correspond to Lettvin?s view. BTW, I talked to Charlie Gross about all this in the 1980s or 90s - I haven?t talked to him about this more recently so his views might have changed. I also knew Horace Barlow a bit, but my conversations with him were about artificial neural nets and computation (again in the 80s and 90s), not neurophysiology. At the first talk I ever gave I notice Horace in the audience and shuddered - he was a bit terrifying to a young neophyte. He did ask a question at the end of my talk, but it wasn?t one that stumped me, and he was very gracious about the whole thing. cheers, Bob Sent from Outlook for Mac From: Asim Roy Date: Friday, 21 August 2026 at 05:16 To: KENTRIDGE, ROBERT W. ; Gary Marcus ; Stephen Jos? Hanson ; Hava Siegelmann (hava.siegelmann at gmail.com) ; Juergen Schmidhuber ; Ali Minai Cc: director at inc.ucsd.edu ; connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] Charlie Gross was in that discussion with Horace Barlow. So was Itzhak Fried and Christof Koch. The referenced Jennifer Aniston experiments were conducted under their guidance at UCLA. And so was Walter Freeman and many other prominent neuro and cognitive scientists involved in that discussion. Gross (2002) characterized grandmother cells as ?multimodal.? Reddy and Thorpe (2014) also define concepts cells as ?invariant, multimodal.? Thus, the fundamental characterization of these two concepts is identical. ?Multimodal invariance? implies associative learning across modalities. They also state that concept cells have ?meaning of a given stimulus in a manner that is invariant to different representations of that stimulus.? ?Selectivity or specificity,? ?complex concept,? ?meaning,? ?multimodal invariance? and ?abstractness? (Reddy and Thorpe, 2014) are all integral properties of both concept cells and grandmother cells. Reddy and Thorpe (2014) categorically state that ?abstract, invariant representations? is ?a hallmark of MTL concept cells.? Quiroga (2012) also admit that concept cells are abstract: ?These and many other examples suggest that MTL neurons encode an abstract representation of the concept triggered by the stimulus.? Unfortunately, Barwich (2019) is completely flawed in its arguments against grandmother cells as argued in Roy?s commentary on the article (Roy (2020)). The neurophysiological evidence for purely abstract single cell representation in the brain points to symbolic representation in the brain. Quiroga et al. (2008) estimate that 40% of MTL cells are abstract. Thus, there is extensive use of abstractions at the single cell level in the brain. This is the neurophysiological evidence for the neurosymbolic approach and what Gary Marcus has been arguing for all along. In a way, both approaches win and need each other. 1. Barlow H. B. (2009). Grandmother cells, symmetry, and invariance: how the term arose and what the facts suggest, in The Cognitive Neurosciences, 4th Edn., ed GazzanigaM. (Cambridge, MA: MIT Press), 309?320. 2. Gross C. G. (2002). Genealogy of the ?grandmother cell?. Neuroscientist8, 512?518. 10.1016/j.meegid.2016.04.006 3. Reddy L. Thorpe S. J. (2014). Concept cells through associative learning of high-level representations. Neuron84, 248?251. 10.1016/j.neuron.2014.10.004 4. Barwich A-S (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 5. Roy A (2020) Commentary: The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 14:59. doi: 10.3389/fnins.2020.00059 6. Quiroga R. Q. Kreiman G. Koch C.FriedI. (2008). Sparse but not ?grandmother-cell? coding in the medial temporal lobe. Trends Cogn. Sci.12, 87?91. 10.1016/j.tics.2007.12.003 Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: KENTRIDGE, ROBERT W. Sent: Thursday, August 20, 2026 3:53 AM To: Asim Roy ; Gary Marcus ; Stephen Jos? Hanson Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... It is worth pointing out that in Jerry Lettvin?s original paper on the grandmother cell the concept was presented as a joke. Lettvin certainly believed in the notion that one might identify cells in the frog?s brain as ?bug detectors?, but he appeared to be sceptical about the idea that such cells might exist in the human brain representing complex concepts like ?my grandmother?. There are a couple of great papers on the history of the grandmother cell: Ann-Sophie Barwich (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 Charles C Gross (2002) Genealogy of the "grandmother cell?. Neuroscientist 8(5):512-8. doi: 10.1177/107385802237175. I remember chatting to Charlie Gross about this around that time (he also reminisced about his part-time job mowing BF Skinner?s lawn). People often assume that the Quiroga et al (2005 Nature, 435. 1102 - 1107) paper on the Jennifer Aniston cell was positive evidence for the existence of grandmother cells in the human brain. This is far from the truth. The Jennifer Aniston cell does not only respond to Jennifer Aniston, but also to related concepts like ?Lisa Kudow? (another actress in ?Friends?), so activity in that cell does not uniquely code ?Jennifer Aniston? and cannot be used to read out the presence of Jennifer Aniston. Quiroga et al make this clear in the discussion of their findings. Bob Kentridge Psychology Department University of Durham Canadian Institute for Advanced Research Programme on Brain, Mind and Consciousness Sent from Outlook for Mac From: Connectionists > on behalf of Asim Roy > Date: Thursday, 20 August 2026 at 06:51 To: Gary Marcus >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] The Horace Barlow note was after a furious private debate with some eminent neuro and cognitive scientists regarding Jennifer Aniston cells being evidence for grandmother cells. And this was at age 93 when he visited me after defending his theory. At that stage of his career, his insights were profound. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Asim Roy Sent: Wednesday, August 19, 2026 6:28 PM To: 'Gary Marcus' >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: FW: AI at 50 videos... I had some communication with Marvin Minsky and Jerry Fodor years ago. Here are two quotes from them: 1. Marvin Minsky: "Don?t pay any attention to the critics. Don?t even ignore them.? -By Sam Goldwyn 2. Jerry Fodor: ?Arguing with Connectionists is like arguing with zombies; both are dead, but neither has noticed it.? And here?s a note from a neuroscience pioneer Horace Barlow, Darwin?s great-grandson (Horace Barlow - Wikipedia), winner of many neuroscience prizes. Without his pioneering work poking into frog brains with microelectrodes, we would NOT have AI as we see it today. The grandmother cell theory essentially points to where cognition exists in an abstract form. And there is plenty of neurophysiological evidence for such single cell abstractions, starting with line orientation cells. ====================================================================================================================================== Dear Asim Yes, I would like to join your group, though I would (again) like to make it clear that the grandmother cell theory is not mine, though I still hold that something conforming amazingly well to what was conceptualised by Jerry Lettvin 50 years ago, really does exist! I am still actively interested in the topic, and have just sent off the final proofs of an essay bearing on it. I shall send you a copy of the final proof as soon as I am allowed to circulate it, and of course the more widely discussed it is the better pleased I shall be, though I fear that what I have written will not be universally accepted, at least at first! Best regards Horace ============================================================================================================= Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Connectionists > On Behalf Of Gary Marcus Sent: Wednesday, August 19, 2026 11:23 AM To: Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... yep: neural networks ascended, pushed symbols out of the way, and then symbols came back relabeled as harnesses, loops code interpreters, and tools. neurosymbolic AI, the hybrid of the two approaches, clearly beat either of the two alternatives on their own. claude code, biggest recent advance, shows that pretty clearly, when you look under the hood. gary On Aug 19, 2026, at 10:16?AM, Stephen Jos? Hanson > wrote: ? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- -- [cid:part1.IzWwLEz4.GBkuYM1Q at rubic.rutgers.edu] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.jpg Type: image/jpeg Size: 57383 bytes Desc: signature.jpg URL: From robert.kentridge at durham.ac.uk Fri Aug 21 13:46:50 2026 From: robert.kentridge at durham.ac.uk (KENTRIDGE, ROBERT W.) Date: Fri, 21 Aug 2026 17:46:50 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Hi Stephen, Your recollections and mine of Charlie?s views seem very similar - Charlie did not believe his face cells and the like were grandmother cells - they were part of a population code rather than being cells whose activity could be read out as unambiguous evidence for the presence of ?x?s. So, his concept must have been closer to my ?hardline? definition than to Reddy and Thorpe?s conception. I knew that Charlie had died in 2019. I think the last time I saw him was at a meeting in Oxford, I don?t remember the date, with Alan Cowey (alas another great, and great friend, we have lost, along with Larry Weiskrantz who was one of my closest friends in visual neuroscience - I published a lot with both of them). At that meeting Charlie was clearly far from well - so sad. cheers, Bob Sent from Outlook for Mac From: Stephen Jos? Hanson Date: Friday, 21 August 2026 at 17:28 To: KENTRIDGE, ROBERT W. ; Asim Roy ; Gary Marcus ; Hava Siegelmann (hava.siegelmann at gmail.com) ; Juergen Schmidhuber ; Ali Minai Cc: director at inc.ucsd.edu ; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] Hi Robert, I was a friend of Charlies in princeton from the 80s. Charlie died in 2019. And we had many dinners together.. occasionally with his most recent wife, Joyce Carol Oats. Charlie argued his face cells (and hand cells0 were not grandmother cells: they were coarsely tuned, responded to many faces, and fit an ensemble/population code. Its hard to fight a counterfactual, as there are too many ways a concept or category intension could be false. How many control conditions do you need before it is categorical response to a specific stimulus? Is an equivalence class enough for it to be categorical? Same problem with so called fuisform face area.. A relevant paper to this is something I wrote a few years ago in Frontiers, Human Neuroscience.. The The Failure of Blobology: fMRI Misinterpretation, Maleficience and Muddle https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2022.870091/full On 8/21/26 12:14, KENTRIDGE, ROBERT W. wrote: I don?t think there is much I disagree with in Asim?s comment. I does go to show, though, how important the details of one?s definition of a grandmother cell is. If you take the hardline definition that a grandmother cell must respond to one concept and one concept alone, so that you can read off the presence of an ?x? from the response of a grandmother cell tuned to ?x?s then the Jennifer Aniston cell is not a grandmother cell. If you take Asim?s definition that they are cells which respond to a complex concept across a range of modalities etc., without any proviso that they must respond exclusively to ?x?s then the Jennifer Aniston cell is indeed a grandmother cell. Reddy and Thorpe certainly describe an interesting class of cell, I?m just not sure if they correspond to Lettvin?s view. BTW, I talked to Charlie Gross about all this in the 1980s or 90s - I haven?t talked to him about this more recently so his views might have changed. I also knew Horace Barlow a bit, but my conversations with him were about artificial neural nets and computation (again in the 80s and 90s), not neurophysiology. At the first talk I ever gave I notice Horace in the audience and shuddered - he was a bit terrifying to a young neophyte. He did ask a question at the end of my talk, but it wasn?t one that stumped me, and he was very gracious about the whole thing. cheers, Bob Sent from Outlook for Mac From: Asim Roy Date: Friday, 21 August 2026 at 05:16 To: KENTRIDGE, ROBERT W. ; Gary Marcus ; Stephen Jos? Hanson ; Hava Siegelmann (hava.siegelmann at gmail.com) ; Juergen Schmidhuber ; Ali Minai Cc: director at inc.ucsd.edu ; connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] Charlie Gross was in that discussion with Horace Barlow. So was Itzhak Fried and Christof Koch. The referenced Jennifer Aniston experiments were conducted under their guidance at UCLA. And so was Walter Freeman and many other prominent neuro and cognitive scientists involved in that discussion. Gross (2002) characterized grandmother cells as ?multimodal.? Reddy and Thorpe (2014) also define concepts cells as ?invariant, multimodal.? Thus, the fundamental characterization of these two concepts is identical. ?Multimodal invariance? implies associative learning across modalities. They also state that concept cells have ?meaning of a given stimulus in a manner that is invariant to different representations of that stimulus.? ?Selectivity or specificity,? ?complex concept,? ?meaning,? ?multimodal invariance? and ?abstractness? (Reddy and Thorpe, 2014) are all integral properties of both concept cells and grandmother cells. Reddy and Thorpe (2014) categorically state that ?abstract, invariant representations? is ?a hallmark of MTL concept cells.? Quiroga (2012) also admit that concept cells are abstract: ?These and many other examples suggest that MTL neurons encode an abstract representation of the concept triggered by the stimulus.? Unfortunately, Barwich (2019) is completely flawed in its arguments against grandmother cells as argued in Roy?s commentary on the article (Roy (2020)). The neurophysiological evidence for purely abstract single cell representation in the brain points to symbolic representation in the brain. Quiroga et al. (2008) estimate that 40% of MTL cells are abstract. Thus, there is extensive use of abstractions at the single cell level in the brain. This is the neurophysiological evidence for the neurosymbolic approach and what Gary Marcus has been arguing for all along. In a way, both approaches win and need each other. 1. Barlow H. B. (2009). Grandmother cells, symmetry, and invariance: how the term arose and what the facts suggest, in The Cognitive Neurosciences, 4th Edn., ed GazzanigaM. (Cambridge, MA: MIT Press), 309?320. 2. Gross C. G. (2002). Genealogy of the ?grandmother cell?. Neuroscientist8, 512?518. 10.1016/j.meegid.2016.04.006 3. Reddy L. Thorpe S. J. (2014). Concept cells through associative learning of high-level representations. Neuron84, 248?251. 10.1016/j.neuron.2014.10.004 4. Barwich A-S (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 5. Roy A (2020) Commentary: The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 14:59. doi: 10.3389/fnins.2020.00059 6. Quiroga R. Q. Kreiman G. Koch C.FriedI. (2008). Sparse but not ?grandmother-cell? coding in the medial temporal lobe. Trends Cogn. Sci.12, 87?91. 10.1016/j.tics.2007.12.003 Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: KENTRIDGE, ROBERT W. Sent: Thursday, August 20, 2026 3:53 AM To: Asim Roy ; Gary Marcus ; Stephen Jos? Hanson Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... It is worth pointing out that in Jerry Lettvin?s original paper on the grandmother cell the concept was presented as a joke. Lettvin certainly believed in the notion that one might identify cells in the frog?s brain as ?bug detectors?, but he appeared to be sceptical about the idea that such cells might exist in the human brain representing complex concepts like ?my grandmother?. There are a couple of great papers on the history of the grandmother cell: Ann-Sophie Barwich (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 Charles C Gross (2002) Genealogy of the "grandmother cell?. Neuroscientist 8(5):512-8. doi: 10.1177/107385802237175. I remember chatting to Charlie Gross about this around that time (he also reminisced about his part-time job mowing BF Skinner?s lawn). People often assume that the Quiroga et al (2005 Nature, 435. 1102 - 1107) paper on the Jennifer Aniston cell was positive evidence for the existence of grandmother cells in the human brain. This is far from the truth. The Jennifer Aniston cell does not only respond to Jennifer Aniston, but also to related concepts like ?Lisa Kudow? (another actress in ?Friends?), so activity in that cell does not uniquely code ?Jennifer Aniston? and cannot be used to read out the presence of Jennifer Aniston. Quiroga et al make this clear in the discussion of their findings. Bob Kentridge Psychology Department University of Durham Canadian Institute for Advanced Research Programme on Brain, Mind and Consciousness Sent from Outlook for Mac From: Connectionists > on behalf of Asim Roy > Date: Thursday, 20 August 2026 at 06:51 To: Gary Marcus >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] The Horace Barlow note was after a furious private debate with some eminent neuro and cognitive scientists regarding Jennifer Aniston cells being evidence for grandmother cells. And this was at age 93 when he visited me after defending his theory. At that stage of his career, his insights were profound. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Asim Roy Sent: Wednesday, August 19, 2026 6:28 PM To: 'Gary Marcus' >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: FW: AI at 50 videos... I had some communication with Marvin Minsky and Jerry Fodor years ago. Here are two quotes from them: 1. Marvin Minsky: "Don?t pay any attention to the critics. Don?t even ignore them.? -By Sam Goldwyn 2. Jerry Fodor: ?Arguing with Connectionists is like arguing with zombies; both are dead, but neither has noticed it.? And here?s a note from a neuroscience pioneer Horace Barlow, Darwin?s great-grandson (Horace Barlow - Wikipedia), winner of many neuroscience prizes. Without his pioneering work poking into frog brains with microelectrodes, we would NOT have AI as we see it today. The grandmother cell theory essentially points to where cognition exists in an abstract form. And there is plenty of neurophysiological evidence for such single cell abstractions, starting with line orientation cells. ====================================================================================================================================== Dear Asim Yes, I would like to join your group, though I would (again) like to make it clear that the grandmother cell theory is not mine, though I still hold that something conforming amazingly well to what was conceptualised by Jerry Lettvin 50 years ago, really does exist! I am still actively interested in the topic, and have just sent off the final proofs of an essay bearing on it. I shall send you a copy of the final proof as soon as I am allowed to circulate it, and of course the more widely discussed it is the better pleased I shall be, though I fear that what I have written will not be universally accepted, at least at first! Best regards Horace ============================================================================================================= Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Connectionists > On Behalf Of Gary Marcus Sent: Wednesday, August 19, 2026 11:23 AM To: Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... yep: neural networks ascended, pushed symbols out of the way, and then symbols came back relabeled as harnesses, loops code interpreters, and tools. neurosymbolic AI, the hybrid of the two approaches, clearly beat either of the two alternatives on their own. claude code, biggest recent advance, shows that pretty clearly, when you look under the hood. gary On Aug 19, 2026, at 10:16?AM, Stephen Jos? Hanson > wrote: ? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- -- [cid:part1.IzWwLEz4.GBkuYM1Q at rubic.rutgers.edu] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.jpg Type: image/jpeg Size: 57383 bytes Desc: signature.jpg URL: From levine at uta.edu Fri Aug 21 11:19:22 2026 From: levine at uta.edu (Levine, Daniel S) Date: Fri, 21 Aug 2026 15:19:22 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: There are more references on grandmother cells in a 2017 special issue of Language, Cognition, and Neuroscience: FOR: Bowers, Language, Cognition, and Neuroscience, 32, 257-273. AGAINST: Thomas and French, Language, Cognition, and Neuroscience, 32, 342-349 These are also discussed in my book, Introduction to Neural and Cognitive Modeling (Vol. III, Routledge, 2019). Dan Levine From: Connectionists On Behalf Of Asim Roy Sent: Thursday, August 20, 2026 11:17 PM To: KENTRIDGE, ROBERT W. ; Gary Marcus ; Stephen Jos? Hanson ; Hava Siegelmann (hava.siegelmann at gmail.com) ; Juergen Schmidhuber ; Ali Minai Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... [External] Charlie Gross was in that discussion with Horace Barlow. So was Itzhak Fried and Christof Koch. The referenced Jennifer Aniston experiments were conducted under their guidance at UCLA. And so was Walter Freeman and many other prominent neuro and cognitive scientists involved in that discussion. Gross (2002) characterized grandmother cells as ?multimodal.? Reddy and Thorpe (2014) also define concepts cells as ?invariant, multimodal.? Thus, the fundamental characterization of these two concepts is identical. ?Multimodal invariance? implies associative learning across modalities. They also state that concept cells have ?meaning of a given stimulus in a manner that is invariant to different representations of that stimulus.? ?Selectivity or specificity,? ?complex concept,? ?meaning,? ?multimodal invariance? and ?abstractness? (Reddy and Thorpe, 2014) are all integral properties of both concept cells and grandmother cells. Reddy and Thorpe (2014) categorically state that ?abstract, invariant representations? is ?a hallmark of MTL concept cells.? Quiroga (2012) also admit that concept cells are abstract: ?These and many other examples suggest that MTL neurons encode an abstract representation of the concept triggered by the stimulus.? Unfortunately, Barwich (2019) is completely flawed in its arguments against grandmother cells as argued in Roy?s commentary on the article (Roy (2020)). The neurophysiological evidence for purely abstract single cell representation in the brain points to symbolic representation in the brain. Quiroga et al. (2008) estimate that 40% of MTL cells are abstract. Thus, there is extensive use of abstractions at the single cell level in the brain. This is the neurophysiological evidence for the neurosymbolic approach and what Gary Marcus has been arguing for all along. In a way, both approaches win and need each other. 1. Barlow H. B. (2009). Grandmother cells, symmetry, and invariance: how the term arose and what the facts suggest, in The Cognitive Neurosciences, 4th Edn., ed GazzanigaM. (Cambridge, MA: MIT Press), 309?320. 2. Gross C. G. (2002). Genealogy of the ?grandmother cell?. Neuroscientist8, 512?518. 10.1016/j.meegid.2016.04.006 3. Reddy L. Thorpe S. J. (2014). Concept cells through associative learning of high-level representations. Neuron84, 248?251. 10.1016/j.neuron.2014.10.004 4. Barwich A-S (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 5. Roy A (2020) Commentary: The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 14:59. doi: 10.3389/fnins.2020.00059 6. Quiroga R. Q. Kreiman G. Koch C.FriedI. (2008). Sparse but not ?grandmother-cell? coding in the medial temporal lobe. Trends Cogn. Sci.12, 87?91. 10.1016/j.tics.2007.12.003 Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: KENTRIDGE, ROBERT W. > Sent: Thursday, August 20, 2026 3:53 AM To: Asim Roy >; Gary Marcus >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... It is worth pointing out that in Jerry Lettvin?s original paper on the grandmother cell the concept was presented as a joke. Lettvin certainly believed in the notion that one might identify cells in the frog?s brain as ?bug detectors?, but he appeared to be sceptical about the idea that such cells might exist in the human brain representing complex concepts like ?my grandmother?. There are a couple of great papers on the history of the grandmother cell: Ann-Sophie Barwich (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 Charles C Gross (2002) Genealogy of the "grandmother cell?. Neuroscientist 8(5):512-8. doi: 10.1177/107385802237175. I remember chatting to Charlie Gross about this around that time (he also reminisced about his part-time job mowing BF Skinner?s lawn). People often assume that the Quiroga et al (2005 Nature, 435. 1102 - 1107) paper on the Jennifer Aniston cell was positive evidence for the existence of grandmother cells in the human brain. This is far from the truth. The Jennifer Aniston cell does not only respond to Jennifer Aniston, but also to related concepts like ?Lisa Kudow? (another actress in ?Friends?), so activity in that cell does not uniquely code ?Jennifer Aniston? and cannot be used to read out the presence of Jennifer Aniston. Quiroga et al make this clear in the discussion of their findings. Bob Kentridge Psychology Department University of Durham Canadian Institute for Advanced Research Programme on Brain, Mind and Consciousness Sent from Outlook for Mac From: Connectionists > on behalf of Asim Roy > Date: Thursday, 20 August 2026 at 06:51 To: Gary Marcus >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] The Horace Barlow note was after a furious private debate with some eminent neuro and cognitive scientists regarding Jennifer Aniston cells being evidence for grandmother cells. And this was at age 93 when he visited me after defending his theory. At that stage of his career, his insights were profound. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Asim Roy Sent: Wednesday, August 19, 2026 6:28 PM To: 'Gary Marcus' >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: FW: AI at 50 videos... I had some communication with Marvin Minsky and Jerry Fodor years ago. Here are two quotes from them: 1. Marvin Minsky: "Don?t pay any attention to the critics. Don?t even ignore them.? -By Sam Goldwyn 2. Jerry Fodor: ?Arguing with Connectionists is like arguing with zombies; both are dead, but neither has noticed it.? And here?s a note from a neuroscience pioneer Horace Barlow, Darwin?s great-grandson (Horace Barlow - Wikipedia), winner of many neuroscience prizes. Without his pioneering work poking into frog brains with microelectrodes, we would NOT have AI as we see it today. The grandmother cell theory essentially points to where cognition exists in an abstract form. And there is plenty of neurophysiological evidence for such single cell abstractions, starting with line orientation cells. ====================================================================================================================================== Dear Asim Yes, I would like to join your group, though I would (again) like to make it clear that the grandmother cell theory is not mine, though I still hold that something conforming amazingly well to what was conceptualised by Jerry Lettvin 50 years ago, really does exist! I am still actively interested in the topic, and have just sent off the final proofs of an essay bearing on it. I shall send you a copy of the final proof as soon as I am allowed to circulate it, and of course the more widely discussed it is the better pleased I shall be, though I fear that what I have written will not be universally accepted, at least at first! Best regards Horace ============================================================================================================= Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Connectionists > On Behalf Of Gary Marcus Sent: Wednesday, August 19, 2026 11:23 AM To: Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... yep: neural networks ascended, pushed symbols out of the way, and then symbols came back relabeled as harnesses, loops code interpreters, and tools. neurosymbolic AI, the hybrid of the two approaches, clearly beat either of the two alternatives on their own. claude code, biggest recent advance, shows that pretty clearly, when you look under the hood. gary On Aug 19, 2026, at 10:16?AM, Stephen Jos? Hanson > wrote: ? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- -------------- next part -------------- An HTML attachment was scrubbed... URL: From gscheler at gmail.com Fri Aug 21 16:26:11 2026 From: gscheler at gmail.com (Gabriele Scheler) Date: Fri, 21 Aug 2026 22:26:11 +0200 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: There is another variant to this, when high-information neurons develop during learning and serve as "index neurons", "concept neurons", linking features to them. https://pubmed.ncbi.nlm.nih.gov/40120001/ 7978335 New Non-standard memory models with indexed retrieval (tomorrow on arxiv) I think fighting over turf is the most damaging aspect of the AI history. The "AI" wars. There is no need to disparage Wiener anymore than the Logic Theorist or the Hebbian/weighted graph tradition. Rigorous ideologically pure "thinking" rarely leads to success in the natural sciences. https://www.researchgate.net/publication/365925517_NIPS_NeurIPS_and_Neuroscience_A_personal_historical_perspective Especially if it is coupled with raw (military) power. I'm with Gary Marcus (and Freud) on noticing that suppression of important parts of a tradition tends to lead to backlashes sooner or later. BTW Hebb was preceded and probably knew about the work of Jerzy Konorski, a former pupil of Pawlow, and a persecuted scientist, BTW. Raw power is rarely successful in the long run. Gabriele On Fri, Aug 21, 2026 at 9:13?PM Levine, Daniel S wrote: > There are more references on grandmother cells in a 2017 special issue of *Language, > Cognition, and Neuroscience*: > > > > > > FOR: Bowers, *Language, Cognition, and Neuroscience*, 32, 257-273. > > AGAINST: Thomas and French, * Language, Cognition, and Neuroscience*, 32, > 342-349 > > > > These are also discussed in my book, *Introduction to Neural and > Cognitive Modeling* (Vol. III, Routledge, 2019). > > > > > > Dan Levine > > *From:* Connectionists *On > Behalf Of *Asim Roy > *Sent:* Thursday, August 20, 2026 11:17 PM > *To:* KENTRIDGE, ROBERT W. ; Gary Marcus < > gary.marcus at nyu.edu>; Stephen Jos? Hanson ; Hava > Siegelmann (hava.siegelmann at gmail.com) ; > Juergen Schmidhuber ; Ali Minai < > minaiaa at gmail.com> > *Cc:* director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu > *Subject:* Re: Connectionists: FW: AI at 50 videos... > > > > [External] > > Charlie Gross was in that discussion with Horace Barlow. So was Itzhak > Fried and Christof Koch. The referenced Jennifer Aniston experiments were > conducted under their guidance at UCLA. And so was Walter Freeman and many > other prominent neuro and cognitive scientists involved in that discussion. > > > > Gross (2002) characterized grandmother cells as ?multimodal.? Reddy and > Thorpe (2014) also define concepts cells as ?invariant, multimodal.? Thus, > the fundamental characterization of these two concepts is identical. > ?Multimodal invariance? implies associative learning across modalities. > They also state that concept cells have ?*meaning* of a given stimulus in > a manner that is *invariant* to different representations of that > stimulus.? > > > > ?*Selectivity or specificity*,? ?*complex concept*,? ?*meaning*,? ?*multimodal > invariance*? and ?*abstractness*? (Reddy and Thorpe, 2014) are all > integral properties of both concept cells and grandmother cells. > > > > Reddy and Thorpe (2014) categorically state that ?*abstract, invariant > representations*? is ?*a hallmark of MTL concept cells*.? Quiroga (2012) > also admit that concept cells are abstract: ?*These and many other > examples suggest that MTL neurons encode an abstract* *representation* of > the concept triggered by the stimulus.? > > > > Unfortunately, Barwich (2019) is completely flawed in its arguments > against grandmother cells as argued in Roy?s commentary on the article (Roy > (2020)). > > > > The neurophysiological evidence for purely abstract single cell > representation in the brain points to symbolic representation in the brain. Quiroga > et al. (2008) estimate that *40% of MTL cells are abstract*. Thus, there > is extensive use of abstractions at the single cell level in the brain. > This is the neurophysiological evidence for the neurosymbolic approach and > what Gary Marcus has been arguing for all along. In a way, both approaches > win and need each other. > > > > 1. Barlow H. B. (2009). Grandmother cells, symmetry, and invariance: > how the term arose and what the facts suggest, in *The Cognitive > Neurosciences, 4th Edn.*, ed GazzanigaM. (Cambridge, MA: MIT > Press), 309?320. > 2. Gross C. G. (2002). Genealogy of the ?grandmother cell?. > *Neuroscientist*8, 512?518. 10.1016/j.meegid.2016.04.006 > > 3. Reddy L. Thorpe S. J. (2014). Concept cells through > associative learning of high-level representations. *Neuron* > 84, 248?251. 10.1016/j.neuron.2014.10.004 > > 4. Barwich A-S (2019) The Value of Failure in Science: The > Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: > 10.3389/fnins.2019.01121 > > 5. Roy A (2020) Commentary: The Value of Failure in Science: > The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 14:59. > doi: 10.3389/fnins.2020.00059 > > 6. Quiroga R. Q. Kreiman G. Koch C.FriedI. (2008). Sparse but > not ?grandmother-cell? coding in the medial temporal lobe. *Trends Cogn. > Sci.*12, 87?91. 10.1016/j.tics.2007.12.003 > > > > Asim Roy > > Professor, Information Systems > > Arizona State University > > Lifeboat Foundation Bios: Professor Asim Roy > > > Asim Roy | iSearch (asu.edu) > > > > > > > > *From:* KENTRIDGE, ROBERT W. > *Sent:* Thursday, August 20, 2026 3:53 AM > *To:* Asim Roy ; Gary Marcus ; > Stephen Jos? Hanson > *Cc:* director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu > *Subject:* Re: Connectionists: FW: AI at 50 videos... > > > > It is worth pointing out that in Jerry Lettvin?s original paper on the > grandmother cell the concept was presented as a joke. Lettvin certainly > believed in the notion that one might identify cells in the frog?s brain as > ?bug detectors?, but he appeared to be sceptical about the idea that such > cells might exist in the human brain representing complex concepts like ?my > grandmother?. There are a couple of great papers on the history of the > grandmother cell: > > > > Ann-Sophie Barwich (2019) The Value of Failure in Science: The Story of > Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: > 10.3389/fnins.2019.01121 > > > > Charles C Gross (2002) Genealogy of the "grandmother cell?. Neuroscientist > 8(5):512-8. doi: 10.1177/107385802237175. > > > > I remember chatting to Charlie Gross about this around that time (he also > reminisced about his part-time job mowing BF Skinner?s lawn). > > > > People often assume that the Quiroga et al (2005 Nature, 435. 1102 - 1107) > paper on the Jennifer Aniston cell was positive evidence for the existence > of grandmother cells in the human brain. This is far from the truth. The > Jennifer Aniston cell does not only respond to Jennifer Aniston, but also > to related concepts like ?Lisa Kudow? (another actress in ?Friends?), so > activity in that cell does not uniquely code ?Jennifer Aniston? and cannot > be used to read out the presence of Jennifer Aniston. Quiroga et al make > this clear in the discussion of their findings. > > > > Bob Kentridge > > > > Psychology Department > > University of Durham > > > > Canadian Institute for Advanced Research > > Programme on Brain, Mind and Consciousness > > > > > > Sent from Outlook for Mac > > > > *From: *Connectionists on > behalf of Asim Roy > *Date: *Thursday, 20 August 2026 at 06:51 > *To: *Gary Marcus ; Stephen Jos? Hanson < > jose at rubic.rutgers.edu> > *Cc: *director at inc.ucsd.edu ; > connectionists at mailman.srv.cs.cmu.edu < > connectionists at mailman.srv.cs.cmu.edu> > *Subject: *Re: Connectionists: FW: AI at 50 videos... > > *[EXTERNAL EMAIL]* > > The Horace Barlow note was after a furious private debate with some > eminent neuro and cognitive scientists regarding Jennifer Aniston cells > being evidence for grandmother cells. And this was at age 93 when he > visited me after defending his theory. At that stage of his career, his > insights were profound. > > > > Asim Roy > > Professor, Information Systems > > Arizona State University > > Lifeboat Foundation Bios: Professor Asim Roy > > > Asim Roy | iSearch (asu.edu) > > > > > > > > *From:* Asim Roy > *Sent:* Wednesday, August 19, 2026 6:28 PM > *To:* 'Gary Marcus' ; Stephen Jos? Hanson < > jose at rubic.rutgers.edu> > *Cc:* director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu > *Subject:* RE: Connectionists: FW: AI at 50 videos... > > > > I had some communication with Marvin Minsky and Jerry Fodor years ago. > Here are two quotes from them: > > > > 1. Marvin Minsky: "Don?t pay any attention to the critics. Don?t even > ignore them.? -By Sam Goldwyn > > > > 2. Jerry Fodor: ?Arguing with Connectionists is like arguing with > zombies; both are dead, but neither has noticed it.? > > > > And here?s a note from a neuroscience pioneer Horace Barlow, Darwin?s > great-grandson (Horace Barlow - Wikipedia > ), winner of many > neuroscience prizes. Without his pioneering work poking into frog brains > with microelectrodes, we would NOT have AI as we see it today. The > grandmother cell theory essentially points to where cognition exists in an > abstract form. And there is plenty of neurophysiological evidence for such > single cell abstractions, starting with line orientation cells. > > > ====================================================================================================================================== > > Dear Asim > > Yes, I would like to join your group, though I would (again) like to make > it clear that the grandmother cell theory is not mine, though I still hold > that something conforming amazingly well to what was conceptualised by > Jerry Lettvin 50 years ago, really does exist! > > I am still actively interested in the topic, and have just sent off the > final proofs of an essay bearing on it. I shall send you a copy of the > final proof as soon as I am allowed to circulate it, and of course the more > widely discussed it is the better pleased I shall be, though I fear that what > I have written will not be universally accepted, at least at first! > > Best regards > > Horace > > > ============================================================================================================= > > > > Asim Roy > > Professor, Information Systems > > Arizona State University > > Lifeboat Foundation Bios: Professor Asim Roy > > > Asim Roy | iSearch (asu.edu) > > > > > > *From:* Connectionists *On > Behalf Of *Gary Marcus > *Sent:* Wednesday, August 19, 2026 11:23 AM > *To:* Stephen Jos? Hanson > *Cc:* director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu > *Subject:* Re: Connectionists: FW: AI at 50 videos... > > > > yep: neural networks ascended, pushed symbols out of the way, and then > symbols came back relabeled as harnesses, loops code interpreters, and > tools. > > > > neurosymbolic AI, the hybrid of the two approaches, clearly beat either of > the two alternatives on their own. > > > > claude code, biggest recent advance, shows that pretty clearly, when you > look under the hood. > > > > gary > > > > On Aug 19, 2026, at 10:16?AM, Stephen Jos? Hanson > wrote: > > ? > > Yes, historically this was an important meeting --and indeed the term AI > was coined by John McCarthy who with several others put together the 1955 > meeting; McCarthy also invented LISP after this meeting. Part of the AI > meeting at DM was political. McCarthy deliberately chose "Artificial > Intelligence" partly to distance the new effort from Wiener's cybernetics > and the neural network tradition. Which was thematic in the MACY meetings.. > which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to > stake out a separate field with a symbolic/logical emphasis (in contrast > to McCulloch and Pitts). So the naming wasn't just descriptive; it was a > bit of a territorial move that arguably sidelined neural networks for > decades until the connectionist revival in the 1980s (sorry Juergen). And > the more recent revolution associated with DL and LLMs more specifically. > In the 1970s, this fueled the long term debate in Cognitive Science on > Symbols/Rules vs Vectors/NN. > > Well at least we know how that turned out. > > Stephen > > > > On 8/17/26 13:30, Terry Sejnowski wrote: > > Jean-Marc - > > I was there and gave a talk, the only one on neural networks. I learned a > lot from the other speakers. > > I just listened to the questions at the end of my talk. I pretty much > predicted what was going to happen in AI and thought it would take 50 > years. I was wrong, it took less than 20 years. > > One of the questions at the end was about intelligent aliens who appear > and don't have human brains. > This was a set piece in my recent book on ChatGPT and the Future of AI, > where ChatGPT is the alien. > > I gave my version of that meeting in my first book on the Deep Learning > Revolution. Every talk that reported progress did so because they had > enough data to analyze, including Gene Charniak who was trying to parse > sentences and Takeo Kanade working on computer vision (These two talks are > not included in the list). At the end Marvin Minsky accused them for just > working on applications, not real AI. > > Terry > > ------ > > On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: > > In case you did not see this: Some amazing lectures and (now historical) > perspectives? the most interesting parts are when they tried and predicted > the future of AI! > > > > *From:* Connectionists > *On Behalf Of *George > Cybenko > *Sent:* Monday, August 17, 2026 6:00 AM > *To:* connectionists at mailman.srv.cs.cmu.edu > *Subject:* Connectionists: AI at 50 videos... > > > > In 2006, we held a 50th anniversary conference at Dartmouth to commemorate > > the historic 1956 meeting that coined the term "artificial intelligence." > At that time, > > five of the surviving original participants attended. Today they are all > gone. > > > > I just recently posted videos of the talks from the AI at 50 meeting at > > > > https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ > > > > > INobody foresaw what is happening in AI today and in fact, > > there were several digs against neural networks etc. > > > > Enjoy! > > > > Regards > > George Cybenko > > -- > > George Cybenko, Dorothy and Walter Gramm Professor of Engineering > > Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA > > > > email: gvc at dartmouth.edu > > FAX: 603 646-9024 > > Cell: 603 369-1133 > > Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu > > > > -- > > > > -------------- next part -------------- An HTML attachment was scrubbed... URL: From ASIM.ROY at asu.edu Sat Aug 22 00:47:08 2026 From: ASIM.ROY at asu.edu (Asim Roy) Date: Sat, 22 Aug 2026 04:47:08 +0000 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: KENTRIDGE, ROBERT W. Sent: Friday, August 21, 2026 10:14 AM To: Asim Roy ; Gary Marcus ; Stephen Jos? Hanson ; Hava Siegelmann (hava.siegelmann at gmail.com) ; Juergen Schmidhuber ; Ali Minai Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... I don?t think there is much I disagree with in Asim?s comment. I does go to show, though, how important the details of one?s definition of a grandmother cell is. If you take the hardline definition that a grandmother cell must respond to one concept and one concept alone, so that you can read off the presence of an ?x? from the response of a grandmother cell tuned to ?x?s then the Jennifer Aniston cell is not a grandmother cell. If you take Asim?s definition that they are cells which respond to a complex concept across a range of modalities etc., without any proviso that they must respond exclusively to ?x?s then the Jennifer Aniston cell is indeed a grandmother cell. Reddy and Thorpe certainly describe an interesting class of cell, I?m just not sure if they correspond to Lettvin?s view. BTW, I talked to Charlie Gross about all this in the 1980s or 90s - I haven?t talked to him about this more recently so his views might have changed. I also knew Horace Barlow a bit, but my conversations with him were about artificial neural nets and computation (again in the 80s and 90s), not neurophysiology. At the first talk I ever gave I notice Horace in the audience and shuddered - he was a bit terrifying to a young neophyte. He did ask a question at the end of my talk, but it wasn?t one that stumped me, and he was very gracious about the whole thing. cheers, Bob Sent from Outlook for Mac From: Asim Roy > Date: Friday, 21 August 2026 at 05:16 To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] Charlie Gross was in that discussion with Horace Barlow. So was Itzhak Fried and Christof Koch. The referenced Jennifer Aniston experiments were conducted under their guidance at UCLA. And so was Walter Freeman and many other prominent neuro and cognitive scientists involved in that discussion. Gross (2002) characterized grandmother cells as ?multimodal.? Reddy and Thorpe (2014) also define concepts cells as ?invariant, multimodal.? Thus, the fundamental characterization of these two concepts is identical. ?Multimodal invariance? implies associative learning across modalities. They also state that concept cells have ?meaning of a given stimulus in a manner that is invariant to different representations of that stimulus.? ?Selectivity or specificity,? ?complex concept,? ?meaning,? ?multimodal invariance? and ?abstractness? (Reddy and Thorpe, 2014) are all integral properties of both concept cells and grandmother cells. Reddy and Thorpe (2014) categorically state that ?abstract, invariant representations? is ?a hallmark of MTL concept cells.? Quiroga (2012) also admit that concept cells are abstract: ?These and many other examples suggest that MTL neurons encode an abstract representation of the concept triggered by the stimulus.? Unfortunately, Barwich (2019) is completely flawed in its arguments against grandmother cells as argued in Roy?s commentary on the article (Roy (2020)). The neurophysiological evidence for purely abstract single cell representation in the brain points to symbolic representation in the brain. Quiroga et al. (2008) estimate that 40% of MTL cells are abstract. Thus, there is extensive use of abstractions at the single cell level in the brain. This is the neurophysiological evidence for the neurosymbolic approach and what Gary Marcus has been arguing for all along. In a way, both approaches win and need each other. 1. Barlow H. B. (2009). Grandmother cells, symmetry, and invariance: how the term arose and what the facts suggest, in The Cognitive Neurosciences, 4th Edn., ed GazzanigaM. (Cambridge, MA: MIT Press), 309?320. 2. Gross C. G. (2002). Genealogy of the ?grandmother cell?. Neuroscientist8, 512?518. 10.1016/j.meegid.2016.04.006 3. Reddy L. Thorpe S. J. (2014). Concept cells through associative learning of high-level representations. Neuron84, 248?251. 10.1016/j.neuron.2014.10.004 4. Barwich A-S (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 5. Roy A (2020) Commentary: The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 14:59. doi: 10.3389/fnins.2020.00059 6. Quiroga R. Q. Kreiman G. Koch C.FriedI. (2008). Sparse but not ?grandmother-cell? coding in the medial temporal lobe. Trends Cogn. Sci.12, 87?91. 10.1016/j.tics.2007.12.003 Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: KENTRIDGE, ROBERT W. > Sent: Thursday, August 20, 2026 3:53 AM To: Asim Roy >; Gary Marcus >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... It is worth pointing out that in Jerry Lettvin?s original paper on the grandmother cell the concept was presented as a joke. Lettvin certainly believed in the notion that one might identify cells in the frog?s brain as ?bug detectors?, but he appeared to be sceptical about the idea that such cells might exist in the human brain representing complex concepts like ?my grandmother?. There are a couple of great papers on the history of the grandmother cell: Ann-Sophie Barwich (2019) The Value of Failure in Science: The Story of Grandmother Cells in Neuroscience. Front. Neurosci. 13:1121. doi: 10.3389/fnins.2019.01121 Charles C Gross (2002) Genealogy of the "grandmother cell?. Neuroscientist 8(5):512-8. doi: 10.1177/107385802237175. I remember chatting to Charlie Gross about this around that time (he also reminisced about his part-time job mowing BF Skinner?s lawn). People often assume that the Quiroga et al (2005 Nature, 435. 1102 - 1107) paper on the Jennifer Aniston cell was positive evidence for the existence of grandmother cells in the human brain. This is far from the truth. The Jennifer Aniston cell does not only respond to Jennifer Aniston, but also to related concepts like ?Lisa Kudow? (another actress in ?Friends?), so activity in that cell does not uniquely code ?Jennifer Aniston? and cannot be used to read out the presence of Jennifer Aniston. Quiroga et al make this clear in the discussion of their findings. Bob Kentridge Psychology Department University of Durham Canadian Institute for Advanced Research Programme on Brain, Mind and Consciousness Sent from Outlook for Mac From: Connectionists > on behalf of Asim Roy > Date: Thursday, 20 August 2026 at 06:51 To: Gary Marcus >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: FW: AI at 50 videos... [EXTERNAL EMAIL] The Horace Barlow note was after a furious private debate with some eminent neuro and cognitive scientists regarding Jennifer Aniston cells being evidence for grandmother cells. And this was at age 93 when he visited me after defending his theory. At that stage of his career, his insights were profound. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Asim Roy Sent: Wednesday, August 19, 2026 6:28 PM To: 'Gary Marcus' >; Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: FW: AI at 50 videos... I had some communication with Marvin Minsky and Jerry Fodor years ago. Here are two quotes from them: 1. Marvin Minsky: "Don?t pay any attention to the critics. Don?t even ignore them.? -By Sam Goldwyn 2. Jerry Fodor: ?Arguing with Connectionists is like arguing with zombies; both are dead, but neither has noticed it.? And here?s a note from a neuroscience pioneer Horace Barlow, Darwin?s great-grandson (Horace Barlow - Wikipedia), winner of many neuroscience prizes. Without his pioneering work poking into frog brains with microelectrodes, we would NOT have AI as we see it today. The grandmother cell theory essentially points to where cognition exists in an abstract form. And there is plenty of neurophysiological evidence for such single cell abstractions, starting with line orientation cells. ====================================================================================================================================== Dear Asim Yes, I would like to join your group, though I would (again) like to make it clear that the grandmother cell theory is not mine, though I still hold that something conforming amazingly well to what was conceptualised by Jerry Lettvin 50 years ago, really does exist! I am still actively interested in the topic, and have just sent off the final proofs of an essay bearing on it. I shall send you a copy of the final proof as soon as I am allowed to circulate it, and of course the more widely discussed it is the better pleased I shall be, though I fear that what I have written will not be universally accepted, at least at first! Best regards Horace ============================================================================================================= Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Connectionists > On Behalf Of Gary Marcus Sent: Wednesday, August 19, 2026 11:23 AM To: Stephen Jos? Hanson > Cc: director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: FW: AI at 50 videos... yep: neural networks ascended, pushed symbols out of the way, and then symbols came back relabeled as harnesses, loops code interpreters, and tools. neurosymbolic AI, the hybrid of the two approaches, clearly beat either of the two alternatives on their own. claude code, biggest recent advance, shows that pretty clearly, when you look under the hood. gary On Aug 19, 2026, at 10:16?AM, Stephen Jos? Hanson > wrote: ? Yes, historically this was an important meeting --and indeed the term AI was coined by John McCarthy who with several others put together the 1955 meeting; McCarthy also invented LISP after this meeting. Part of the AI meeting at DM was political. McCarthy deliberately chose "Artificial Intelligence" partly to distance the new effort from Wiener's cybernetics and the neural network tradition. Which was thematic in the MACY meetings.. which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to stake out a separate field with a symbolic/logical emphasis (in contrast to McCulloch and Pitts). So the naming wasn't just descriptive; it was a bit of a territorial move that arguably sidelined neural networks for decades until the connectionist revival in the 1980s (sorry Juergen). And the more recent revolution associated with DL and LLMs more specifically. In the 1970s, this fueled the long term debate in Cognitive Science on Symbols/Rules vs Vectors/NN. Well at least we know how that turned out. Stephen On 8/17/26 13:30, Terry Sejnowski wrote: Jean-Marc - I was there and gave a talk, the only one on neural networks. I learned a lot from the other speakers. I just listened to the questions at the end of my talk. I pretty much predicted what was going to happen in AI and thought it would take 50 years. I was wrong, it took less than 20 years. One of the questions at the end was about intelligent aliens who appear and don't have human brains. This was a set piece in my recent book on ChatGPT and the Future of AI, where ChatGPT is the alien. I gave my version of that meeting in my first book on the Deep Learning Revolution. Every talk that reported progress did so because they had enough data to analyze, including Gene Charniak who was trying to parse sentences and Takeo Kanade working on computer vision (These two talks are not included in the list). At the end Marvin Minsky accused them for just working on applications, not real AI. Terry ------ On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: In case you did not see this: Some amazing lectures and (now historical) perspectives? the most interesting parts are when they tried and predicted the future of AI! From: Connectionists On Behalf Of George Cybenko Sent: Monday, August 17, 2026 6:00 AM To: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... In 2006, we held a 50th anniversary conference at Dartmouth to commemorate the historic 1956 meeting that coined the term "artificial intelligence." At that time, five of the surviving original participants attended. Today they are all gone. I just recently posted videos of the talks from the AI at 50 meeting at https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ INobody foresaw what is happening in AI today and in fact, there were several digs against neural networks etc. Enjoy! Regards George Cybenko -- George Cybenko, Dorothy and Walter Gramm Professor of Engineering Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA email: gvc at dartmouth.edu FAX: 603 646-9024 Cell: 603 369-1133 Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu -- -------------- next part -------------- An HTML attachment was scrubbed... URL: From icaitemspisa at gmail.com Sat Aug 22 04:42:49 2026 From: icaitemspisa at gmail.com (ICAI TEMS 2026) Date: Sat, 22 Aug 2026 10:42:49 +0200 Subject: Connectionists: [CFP] IEEE ICAI-TEMS 2026 full papers due 25 August (strict) Message-ID: <52cd73b4-acb6-41f3-ad9f-10add2f85c2a@gmail.com> Dear Colleagues, *A final extension has been granted for IEEE ICAI-TEMS 2026: full papers are now due on 25 August 2026. This deadline is strict and will not be extended again.* IEEE ICAI-TEMS 2026 Pisa, Italy, 10-13 November 2026, sponsored by IEEE TEMS. Accepted and presented papers will be submitted for inclusion in IEEE Xplore and Scopus. ? *Full paper submission: 25 August 2026 (strict)* ? Notification of acceptance: 11 September 2026 ? Camera-ready: 25 September 2026 Tracks, author guidelines and submission: https://www.icai-tems.eu/call_for_paper.html If your manuscript is nearly ready, this is the moment to finalise it and please forward this message to colleagues and PhD students who may be interested. For more informations visit the website www.icai-tems.eu * For any other questions, feel free to contact the organizing committee at icai.tems.eu at gmail.com . * Best regards, Pietro Calabrese, Publicity Chair, IEEE ICAI-TEMS 2026 -------------- next part -------------- An HTML attachment was scrubbed... URL: From suashdeb at gmail.com Mon Aug 24 03:05:03 2026 From: suashdeb at gmail.com (Suash Deb) Date: Mon, 24 Aug 2026 12:35:03 +0530 Subject: Connectionists: ISMSI 2027 submission deadline 3 months away Message-ID: Dear Esteemed Colleagues, This is to bring to your attention abt ISMSI 2027 , the 11th edition of the annual international event of IICCI http://www.iicci.in/ The deadline for submission of manuscripts of the above is 3 months away https://www.ismsi.org/ I solicit your support in sharing this info among the peers in your network and motivate them to submit their manuscripts for this conference. Will look forward to receiving quality manuscripts from your peers in the coming days. Thanks very much in advance and with kind rgds, Suash Suash Deb General Chair, ISMSI 2027 -------------- next part -------------- An HTML attachment was scrubbed... URL: From frothga at sandia.gov Mon Aug 24 11:46:10 2026 From: frothga at sandia.gov (Rothganger, Fred) Date: Mon, 24 Aug 2026 15:46:10 +0000 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists on behalf of Asim Roy Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. ; Gary Marcus ; Stephen Jos? Hanson ; Hava Siegelmann (hava.siegelmann at gmail.com) ; Juergen Schmidhuber ; Ali Minai Cc: director at inc.ucsd.edu ; connectionists at mailman.srv.cs.cmu.edu Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -------------- next part -------------- An HTML attachment was scrubbed... URL: From zk240 at cam.ac.uk Mon Aug 24 14:19:01 2026 From: zk240 at cam.ac.uk (Zoe Kourtzi) Date: Mon, 24 Aug 2026 18:19:01 +0000 Subject: Connectionists: Postdoc position in Neurotechnology, Univ of Cambridge In-Reply-To: <54745CB4-56B9-45E5-926E-CC4AFC4FC046@cam.ac.uk> References: <54745CB4-56B9-45E5-926E-CC4AFC4FC046@cam.ac.uk> Message-ID: Interested in translating cognitive neuroscience to neurotechnology solutions? Join the Adaptive Brain Lab at the University of Cambridge as a Post-doctoral Research Associate in Neurotechnology. Work with a cross-disciplinary international team of leading experts investigating brain network plasticity, multimodal neuroimaging and neuromodulation for adaptive behaviour. Apply by 1 September 2026: https://www.cam.ac.uk/jobs/research-assistantassociate-pj50742 For Informal enquiries please contact Prof Zoe Kourtzi (zk240 at cam.ac.uk) with CV and brief statement of background skills and research interests. -------------- next part -------------- An HTML attachment was scrubbed... URL: From ASIM.ROY at asu.edu Mon Aug 24 22:56:18 2026 From: ASIM.ROY at asu.edu (Asim Roy) Date: Tue, 25 Aug 2026 02:56:18 +0000 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: * Plate (2002): "Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995)." * Thorpe (1995, p. 550): "With a local representation, activity in individual units can be interpreted directly ... with distributed coding individual units cannot be interpreted without knowing the state of other units in the network." * Elman (1995, p. 210): "These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units." 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195-223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and "meaning." Reddy and Thorpe (2014) conclude: "In conclusion, evidence is accumulating that "concept cells" carry high-level, abstract stimulus information." Single cells having "meaning" was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had "meaning." "Meaning" and abstraction is at the heart of these debates, not grandmother cells. And finding "meaning" in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -------------- next part -------------- An HTML attachment was scrubbed... URL: From aurel at ee.columbia.edu Mon Aug 24 15:03:51 2026 From: aurel at ee.columbia.edu (Aurel A. Lazar) Date: Mon, 24 Aug 2026 15:03:51 -0400 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: <1E32953D-2E2F-47F2-8426-330620B9513B@ee.columbia.edu> Fred, The notion of ?meaning? (semantics) used to be unquestionably an ill-defined term in the neuroscience literature. However, with the case of olfaction, I believe that to no longer be true. Odorant semantics is now formally defined as odorant identity. Please see the review paper below: Aurel A. Lazar and Yiyin Zhou The Connectome and the Quest for the Functional Logic of the Drosophila Early Olfactory System arxiv , August 2026 BibTex?? DOI Happy to answer any questions that you may have. Best, Aurel http://www.bionet.ee.columbia.edu > On Aug 24, 2026, at 11:46?AM, Rothganger, Fred wrote: > > Be Careful With This Message > The sender's identity could not be verified and someone may be impersonating the sender. > Asim, > > It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. > > -- Fred > > > From: Connectionists > on behalf of Asim Roy > > Sent: Friday, August 21, 2026 10:47 PM > To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com ) >; Juergen Schmidhuber >; Ali Minai > > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... > > By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: > > ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? > > Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? > > ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. > > Asim Roy > Professor, Information Systems > Arizona State University > Lifeboat Foundation Bios: Professor Asim Roy > Asim Roy | iSearch (asu.edu) -------------- next part -------------- An HTML attachment was scrubbed... URL: From steve at bu.edu Mon Aug 24 14:59:03 2026 From: steve at bu.edu (Grossberg, Stephen) Date: Mon, 24 Aug 2026 18:59:03 +0000 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: My new book Your Creative Brain and AI: How We Learn and Consciously Experience Art, Music, and Meaning develops a neural model of how children learn language in real time from their caregivers as well as the perceptual and affective meanings that the linguistic utterances describe. You can find it on Amazon for very little money. I subsidized the cost with my personal funds. Get Outlook for iOS ________________________________ From: Connectionists on behalf of Rothganger, Fred Sent: Monday, 24 August 2026 11:46:10 To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Connectionists: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists on behalf of Asim Roy Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. ; Gary Marcus ; Stephen Jos? Hanson ; Hava Siegelmann (hava.siegelmann at gmail.com) ; Juergen Schmidhuber ; Ali Minai Cc: director at inc.ucsd.edu ; connectionists at mailman.srv.cs.cmu.edu Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -------------- next part -------------- An HTML attachment was scrubbed... URL: From ASIM.ROY at asu.edu Mon Aug 24 23:16:35 2026 From: ASIM.ROY at asu.edu (Asim Roy) Date: Tue, 25 Aug 2026 03:16:35 +0000 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Although we argued a lot, Walter Freeman and I were good friends. Asim From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and "meaning." Reddy and Thorpe (2014) conclude: "In conclusion, evidence is accumulating that "concept cells" carry high-level, abstract stimulus information." Single cells having "meaning" was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had "meaning." "Meaning" and abstraction is at the heart of these debates, not grandmother cells. And finding "meaning" in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -------------- next part -------------- An HTML attachment was scrubbed... URL: From laurenz.wiskott at rub.de Tue Aug 25 04:35:36 2026 From: laurenz.wiskott at rub.de (Laurenz Wiskott) Date: Tue, 25 Aug 2026 10:35:36 +0200 Subject: Connectionists: 2 PhD positions for 4 years with focus on "Generative Episodic Memory" with Prof. Laurenz Wiskott, Germany, Bochum Message-ID: >From https://jobs.ruhr-uni-bochum.de/jobposting/854eabcb1786fa8b53af20c255cd9aeadb5dfc4e1?ref=homepage Faculty of Computer Science : Chair: Theory of Neural Systems In order to fill a fixed-term position in part-time (29.87 hours/week = 75%) at the earliest possible date, we are looking for 2 2 Research associates (m/f/x) for 4 years with focus on "Generative Episodic Memory" The Institute for Neural Computation is a central research institute at the Ruhr-University Bochum, see https://www.ini.rub.de/. It focuses on the dynamics and learning of perception and behavior on a functional level but is otherwise very diverse, ranging from neurophysiology and psychophysics over computational neuroscience to machine learning and technical applications. The research focuses on (i) modeling generative episodic memory in close collaboration with experimental partners, with the goal of modeling experimental data and gaining a deeper understanding of memory processes; (ii) developing a hierarchical model of the visual system that supports a generative memory process. The first project will focus primarily on standard machine learning methods, such as Transformers. The second project aims to develop new methods inspired by the brain. Scope: part-time (75%) Duration: fixed-term, 4 Jahre Start: at the earliest possible date Apply by: 2026-09-14 Your tasks: (i) Further development of an existing system model of generative episodic memory in light of various experimental results. (ii) Development of a brain-inspired hierarchical model of the visual system that supports distributed generative memory processes. Modeling of experimental data. Communication with experimental partners. Publication of results. Teaching assistance equivalent to 3 SWS, in particular conducting a Python course and supervising student/Bachelor/Master projects. Your profile: Requirements: Very good Master degree in mathematics, computer science, engineering, or a related field. Interest in interdisciplinary research in the area of neuroscience. Good programming and mathematical skills. Basic knowledge in machine learning. Advantageous: Experience in modeling. Experience in interdisciplinary projects. Programming skills in Python. We offer: Challenging and varied tasks with a high level of independence Collaboration in a committed and appreciative team Extensive training and professional development opportunities An interesting interdisciplinary environment in the field of theoretical brain research / machine learning. Freedom to shape the project yourself. Infrastructure that facilitates close integration into the institute and the research group. Opportunity to pursue a Ph.D. Further information: The position is salaried and based on the collective agreement of the L?nder (TV-L). If the personal and collective agreement requirements are met, the employee will receive pay grade E 13 TV-L. Further information can be found at https://oeffentlicher-dienst.info/ (in German). The place of work is Ruhr University Bochum. The load of teaching will be calculated according to ? 3 of Lehrverpflichtungsverordnung (state of North Rhine-Westphalia). Applications (CV, transcript of records for MSc and Bsc, letter of motivation) should be sent as a single pdf file. RUB sees itself as a university with an international presence. The campus languages are German and English. Competence in at least one of the two languages and the willingness to learn the other are a prerequisite. RUB provides corresponding free courses for employees. German language courses are offered by the University Language Center (ZFA) in the field of German as a Foreign Language (DaF). https://www.daf.ruhr-uni-bochum.de/daf/mitarbeitende/index.html.en The Staff Council has the right to participate in all selection interviews. At the request of a candidate (m/f/x), it will ensure its participation in the entire procedure. Please contact wpr at rub.de. The Ruhr-Universit?t Bochum is one of Germany?s leading research universities, addressing the whole range of academic disciplines. A highly dynamic setting enables researchers and students to work across the traditional boundaries of academic subjects and faculties. To create knowledge networks within and beyond the university is Ruhr-Universit?t Bochum?s declared aim. The Ruhr-Universit?t Bochum stands for diversity and equal opportunities. For this reason, we favour a working environment composed of heterogeneous teams, and seek to promote the careers of individuals who are underrepresented in our respective professional areas. The Ruhr-Universit?t Bochum expressly requests job applications from women. In areas in which they are underrepresented they will be given preference in the case of equivalent qualifications with male candidates. Applications from individuals with disabilities are most welcome. Contact persons for further information: Prof. Dr. Laurenz Wiskott Tel.: +49 234 32 27997 Kathleen Schmidt Tel.: +49 234 32 27051 Travel costs, accommodation costs and loss of earnings or other application costs for job interviews can unfortunately not be reimbursed. We look forward to receiving your application via our online application portal at https://jobs.ruhr-uni-bochum.de/en/jobposting/854eabcb1786fa8b53af20c255cd9aeadb5dfc4e1/apply?ref=homepage by 2026-09-14. Please make sure to mention the reference number ANR 6046. From claudio.piciarelli at uniud.it Tue Aug 25 08:33:05 2026 From: claudio.piciarelli at uniud.it (Claudio Piciarelli) Date: Tue, 25 Aug 2026 12:33:05 +0000 Subject: Connectionists: Call for Papers: VisionDocs @ WACV 2027 In-Reply-To: References: Message-ID: [Apologies if you receive multiple copies] =================================== CALL FOR PAPERS: VisionDocs @ WACV 2027 4th Workshop on Computer Vision Systems for Document Analysis and Recognition Website: https://sites.google.com/view/avml-lab-visiondocs-wacv2027/ =================================== We are pleased to announce VisionDocs: 4th Workshop on Computer Vision Systems for Document Analysis and Recognition, held in conjunction with the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2027, at Disney Springs, Buena Vista, Orlando, Florida, USA, on January 4 or 5, 2027. Workshop Overview The rapid growth of foundation models and multimodal AI has transformed document understanding into a key research area with broad scientific and industrial relevance. VisionDocs aims to bring together researchers working on computer vision, multimodal learning, and document intelligence to discuss the next generation of AI systems for document understanding. We particularly welcome contributions on vision-language foundation models, agentic document AI, generative approaches, efficient adaptation, and robust and trustworthy document intelligence. Call for Papers Research papers are solicited in, but not limited to, the following topic areas: * Document image processing * Physical and logical layout analysis * Text and symbol recognition * Handwriting recognition * Document analysis systems * Document classification * Recognition of tables and formulas * Data-efficient document analysis * Document synthesis * Document visual question answering * Extracting document semantics * Structured document generation * Historical document analysis * Agentic systems for document understanding * Multi-modal document analysis * Datasets and benchmarks for document analysis Submission We invite submissions of original and unpublished work related to the workshop themes: * Regular papers: up to 8 pages + references. Accepted regular papers will be published in the WACV 2027 Workshop Proceedings. * Short papers: up to 4 pages + references. Accepted short papers will be published on the VisionDocs website only and will not be included in the WACV proceedings. All submissions must follow the WACV 2027 formatting guidelines and be prepared for double-blind review. Accepted papers must be presented in person during the workshop, either as oral presentations or posters. Important Dates Regular Papers * Paper submission: October 13, 2026, 11:59 PM Pacific Time * Author notification: November 2, 2026, 11:59 PM Pacific Time * Camera-ready: November 20, 2026, 11:59 PM Pacific Time Short Papers * Paper submission: November 12, 2026, 11:59 PM Pacific Time * Author notification: November 19, 2026, 11:59 PM Pacific Time * Camera-ready: November 26, 2026, 11:59 PM Pacific Time Workshop: January 4 or 5, 2027 For additional information and submission details, please visit: https://sites.google.com/view/avml-lab-visiondocs-wacv2027/ Contact: visiondocs.organizers at gmail.com The VisionDocs Organizers Silvia Zottin, Axel De Nardin, Silvia Cascianelli, Claudio Piciarelli, and Gian Luca Foresti -------------- next part -------------- An HTML attachment was scrubbed... URL: From emergingtechnetwork.publicity at gmail.com Tue Aug 25 09:23:39 2026 From: emergingtechnetwork.publicity at gmail.com (=?UTF-8?Q?Gizem_G=C3=BCltekin_Varkonyi?=) Date: Tue, 25 Aug 2026 16:23:39 +0300 Subject: Connectionists: DTISO CFP: The IEEE co-sponsored International Conference on Digital Transformation and Intelligent SOciety, 17 - 20 November 2026 | Barcelona, Spain Message-ID: [Apologies if you got multiple copies of this invitation] The International Conference on Digital Transformation and Intelligent SOciety (DTISO 2026) https://dtis-conference.org/2026/ 17 - 20 November 2026 | Barcelona, Spain Hybrid Event and Technically Co-Sponsored by IEEE Spain Section *DTISO 2026 CFP:* Digital transformation and intelligent society refer to the profound integration of advanced digital technologies?such as artificial intelligence, big data analytics, cloud computing, and the Internet of Things into every aspect of modern life. This transformation is reshaping how individuals communicate, how organizations operate, and how governments deliver services. By leveraging data-driven insights and intelligent systems, digital transformation enhances efficiency, enables automation, and supports more informed and timely decision-making across sectors including healthcare, education, industry, and public administration. An intelligent society builds upon these capabilities to create interconnected, adaptive, and human-centered environments where technology is used not only to optimize processes but also to improve quality of life, promote inclusivity, and support sustainable development. As digital ecosystems continue to evolve, the synergy between technological innovation and societal needs becomes essential in addressing global challenges and driving long-term economic and social progress. The International Conference on Digital Transformation and Intelligent Society (DTISO) invites researchers, practitioners, and industry experts to submit their original and high-quality contributions addressing the challenges and opportunities of digital transformation in shaping an intelligent society. We invite the submission of original papers on all topics related to Intelligent Systems for Cybersecurity, with special interest in but not limited to: *?** Digital Transformation Foundations * - Digital Transformation innovation and strategy - Smart cities and smart environments - Industry 4.0 and 5.0 - Digital governance and e-government - Digital economy and business transformation *?** Artificial Intelligence for Digital Society * - Generative AI and Large Language Models - Intelligent Automation and Robotics - Natural Language Processing and Conversational Systems - Computer Vision and Multimedia Intelligence - Predictive Analytics and Intelligent Forecasting - Multi-Agent and Autonomous Systems - Recommender Systems and Personalization - Knowledge Graphs and Semantic Technologies - AI-Driven Decision Support Systems *?** Digital Infrastructure and Emerging Technologies * - Cloud Computing and Digital Platforms - Edge Computing and Intelligent Infrastructure - Internet of Things (IoT) and Connected Systems - 5G/6G and Future Communication Systems - Blockchain and Decentralized Applications - Quantum Computing and Future Digital Technologies - Digital Twin Technologies - Cybersecurity and Digital Trust - Privacy, Ethics, and Responsible AI *?** Smart Society and Human-Centered Systems * - Intelligent Society and Smart Living Technologies - Smart Cities and Urban Intelligence - E-Government and Digital Public Services - Smart Education and E-Learning Technologies - Intelligent Healthcare and Digital Health Systems - Social Computing and Digital Communities - Human-Centered AI and Intelligent Interaction - Digital Inclusion and Accessibility - Ethical and Societal Impact of Emerging Technologies - Digital Media, Misinformation, and Content Intelligence *?** Industry, Economy, and Sustainable Transformation* - Digital Transformation Strategies and Frameworks - Digital Economy and Smart Business Models - Industry 4.0 and Smart Manufacturing - FinTech and Intelligent Financial Systems - Smart Agriculture and Precision Farming - Intelligent Transportation and Mobility Systems - Sustainable Digital Transformation - Smart Energy and Green Technologies - AI Governance and Regulatory Technologies *?** Data Science and Intelligent Information Systems* - Big Data Analytics for Smart Society - Intelligent Information Retrieval and Search Systems - Data-Driven Smart Applications - Advanced Data Processing and Analytics - Real-Time Monitoring and Intelligent Systems - Context-Aware and Adaptive Systems - Data Privacy and Secure Information Systems - Intelligent Decision Analytics - Semantic and Cognitive Computing Systems * Submission Types:* Accepted types of submissions are including: * Full papers (8 Pages), * Short papers and PhD forum (6 Pages) * Workshop papers (6 Pages) * Posters and demos (2 Pages). All accepted papers will be published in the conference proceedings and submitted to IEEE for publication and to be indexed by Scopus. *Submissions Guidelines and Proceedings* Manuscripts should be prepared in 10-point font using the IEEE 8.5" x 11" two-column format. All papers should be in PDF format, and submitted electronically at Paper Submission Link. A full paper can be up to 8 pages (including all figures, tables and references). Submitted papers must present original unpublished research that is not currently under review for any other conference or journal. Papers not following these guidelines may be rejected without review. Also submissions received after the due date, exceeding length limit, or not appropriately structured may also not be considered. Authors may contact the Program Chair for further information or clarification. All submissions are peer-reviewed by at least three reviewers. Accepted papers will appear in the DTISO Proceeding, and be published by the IEEE Computer Society Conference Publishing Services and be submitted to IEEE Xplore for inclusion. *Important Dates:* - Paper Submission: *September 10, 2026 * *(Firm and Final) * - Notification to Authors: October 5, 2026 - Camera Ready Submission: October 20, 2026 *Contact:* Please send any inquiry on DTISO to : info at dtis-conference.org -------------- next part -------------- An HTML attachment was scrubbed... URL: From barak at pearlmutter.net Tue Aug 25 11:21:40 2026 From: barak at pearlmutter.net (Barak A. Pearlmutter) Date: Tue, 25 Aug 2026 16:21:40 +0100 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: > Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? > Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? > Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? People blithely toss out statements like that, but local uninterpretability does not follow from the definition of a distributed representation. It is very easy to construct distributed representations whose components *can* be understood in isolation (the bits of a standard binary representation of a natural number ?2??, say) or whose components cannot be understood in isolation (a code for a natural number ?2?? whose bits each have an associated set of half the numbers in the domain picked at random, say). The "big regions" distributed representation for points in 2D space from the PDP books is locally interpretable. Naturally, all other things being equal, given the choice we'd choose distributed representations whose parts are interpretable in isolation. Does the brain do that? Well, deep learning networks seem to, and experimental data seems consistent with that notion. --Barak Pearlmutter. From lorenzr at cbs.mpg.de Wed Aug 26 05:15:31 2026 From: lorenzr at cbs.mpg.de (Romy Lorenz) Date: Wed, 26 Aug 2026 11:15:31 +0200 Subject: Connectionists: =?utf-8?q?PhD_Position_in_Ultra-High-Field_Layer_?= =?utf-8?q?fMRI_of_Cognitive_Control_and_Interoception_=40_MPI_T=C3=BCbing?= =?utf-8?q?en?= Message-ID: <614E8872-30C5-4715-8D72-9A1564845018@cbs.mpg.de> My lab at the MPI for Biological Cybernetics in T?bingen, Germany is currently recruiting a new PhD student for an exciting cross-species collaborative project using layer fMRI to investigate how interoception shapes cognitive control. Please feel free to share with your colleagues! Application deadline is: 1.10.2026 Start: Jan 2027 More info: https://www.mpg.de/26934670/phd-position-ultra-high-field-layer-fmri-of-cognitive-control-and-interoception If you have questions about this position, please reach out! Best, Romy Dr. Romy Lorenz Max Planck Research Group Leader Research Group Cognitive Neuroscience & Neurotechnology Max Planck Institute for Biological Cybernetics T?bingen, Germany romy.lorenz at tuebingen.mpg.de www.kyb.tuebingen.mpg.de/711763/cognitive-neuroscience-neurotechnology -------------- next part -------------- An HTML attachment was scrubbed... URL: From hava.siegelmann at gmail.com Wed Aug 26 07:36:42 2026 From: hava.siegelmann at gmail.com (Hava Siegelmann) Date: Wed, 26 Aug 2026 14:36:42 +0300 Subject: Connectionists: FW: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: It was Minsky himself who said he was the devil. I was actually at the AI50 and heard the repeat of it :-) On Fri, Aug 21, 2026 at 8:59?AM Asim Roy wrote: > Hava, you are not dead. Those statements of Marvin Minsky and Jerry Fodor > reflect the bitterness of those days. I have also heard a connectionist > call Marvin a ?devil? in an acceptance speech for an award. > > > > Asim Roy > > Professor, Information Systems > > Arizona State University > > Lifeboat Foundation Bios: Professor Asim Roy > > > Asim Roy | iSearch (asu.edu) > > > > > > *From:* Hava Siegelmann > *Sent:* Thursday, August 20, 2026 8:29 AM > *To:* Asim Roy > *Cc:* Gary Marcus ; Stephen Jos? Hanson < > jose at rubic.rutgers.edu>; director at inc.ucsd.edu; > connectionists at mailman.srv.cs.cmu.edu > *Subject:* Re: Connectionists: FW: AI at 50 videos... > > > > This is amazing I love it - dead ! > > > > On Thu, Aug 20, 2026 at 8:30?AM Asim Roy wrote: > > I had some communication with Marvin Minsky and Jerry Fodor years ago. > Here are two quotes from them: > > > > 1. Marvin Minsky: "Don?t pay any attention to the critics. Don?t even > ignore them.? -By Sam Goldwyn > > > > 2. Jerry Fodor: ?Arguing with Connectionists is like arguing with > zombies; both are dead, but neither has noticed it.? > > > > And here?s a note from a neuroscience pioneer Horace Barlow, Darwin?s > great-grandson (Horace Barlow - Wikipedia > ), winner of many > neuroscience prizes. Without his pioneering work poking into frog brains > with microelectrodes, we would NOT have AI as we see it today. The > grandmother cell theory essentially points to where cognition exists in an > abstract form. And there is plenty of neurophysiological evidence for such > single cell abstractions, starting with line orientation cells. > > > ====================================================================================================================================== > > Dear Asim > > Yes, I would like to join your group, though I would (again) like to make > it clear that the grandmother cell theory is not mine, though I still hold > that something conforming amazingly well to what was conceptualised by > Jerry Lettvin 50 years ago, really does exist! > > I am still actively interested in the topic, and have just sent off the > final proofs of an essay bearing on it. I shall send you a copy of the > final proof as soon as I am allowed to circulate it, and of course the more > widely discussed it is the better pleased I shall be, though I fear that what > I have written will not be universally accepted, at least at first! > > Best regards > > Horace > > > ============================================================================================================= > > > > Asim Roy > > Professor, Information Systems > > Arizona State University > > Lifeboat Foundation Bios: Professor Asim Roy > > > Asim Roy | iSearch (asu.edu) > > > > > > *From:* Connectionists *On > Behalf Of *Gary Marcus > *Sent:* Wednesday, August 19, 2026 11:23 AM > *To:* Stephen Jos? Hanson > *Cc:* director at inc.ucsd.edu; connectionists at mailman.srv.cs.cmu.edu > *Subject:* Re: Connectionists: FW: AI at 50 videos... > > > > yep: neural networks ascended, pushed symbols out of the way, and then > symbols came back relabeled as harnesses, loops code interpreters, and > tools. > > > > neurosymbolic AI, the hybrid of the two approaches, clearly beat either of > the two alternatives on their own. > > > > claude code, biggest recent advance, shows that pretty clearly, when you > look under the hood. > > > > gary > > > > On Aug 19, 2026, at 10:16?AM, Stephen Jos? Hanson > wrote: > > ? > > Yes, historically this was an important meeting --and indeed the term AI > was coined by John McCarthy who with several others put together the 1955 > meeting; McCarthy also invented LISP after this meeting. Part of the AI > meeting at DM was political. McCarthy deliberately chose "Artificial > Intelligence" partly to distance the new effort from Wiener's cybernetics > and the neural network tradition. Which was thematic in the MACY meetings.. > which ended in 1953--2 years before the "AI" meeting. McCarthy wanted to > stake out a separate field with a symbolic/logical emphasis (in contrast > to McCulloch and Pitts). So the naming wasn't just descriptive; it was a > bit of a territorial move that arguably sidelined neural networks for > decades until the connectionist revival in the 1980s (sorry Juergen). And > the more recent revolution associated with DL and LLMs more specifically. > In the 1970s, this fueled the long term debate in Cognitive Science on > Symbols/Rules vs Vectors/NN. > > Well at least we know how that turned out. > > Stephen > > > > On 8/17/26 13:30, Terry Sejnowski wrote: > > Jean-Marc - > > I was there and gave a talk, the only one on neural networks. I learned a > lot from the other speakers. > > I just listened to the questions at the end of my talk. I pretty much > predicted what was going to happen in AI and thought it would take 50 > years. I was wrong, it took less than 20 years. > > One of the questions at the end was about intelligent aliens who appear > and don't have human brains. > This was a set piece in my recent book on ChatGPT and the Future of AI, > where ChatGPT is the alien. > > I gave my version of that meeting in my first book on the Deep Learning > Revolution. Every talk that reported progress did so because they had > enough data to analyze, including Gene Charniak who was trying to parse > sentences and Takeo Kanade working on computer vision (These two talks are > not included in the list). At the end Marvin Minsky accused them for just > working on applications, not real AI. > > Terry > > ------ > > On 8/17/2026 9:18 AM, Fellous, Jean-Marc wrote: > > In case you did not see this: Some amazing lectures and (now historical) > perspectives? the most interesting parts are when they tried and predicted > the future of AI! > > > > *From:* Connectionists > *On Behalf Of *George > Cybenko > *Sent:* Monday, August 17, 2026 6:00 AM > *To:* connectionists at mailman.srv.cs.cmu.edu > *Subject:* Connectionists: AI at 50 videos... > > > > In 2006, we held a 50th anniversary conference at Dartmouth to commemorate > > the historic 1956 meeting that coined the term "artificial intelligence." > At that time, > > five of the surviving original participants attended. Today they are all > gone. > > > > I just recently posted videos of the talks from the AI at 50 meeting at > > > > https://sites.dartmouth.edu/cybenko/dartmouth-ai50-videos/ > > > > > INobody foresaw what is happening in AI today and in fact, > > there were several digs against neural networks etc. > > > > Enjoy! > > > > Regards > > George Cybenko > > -- > > George Cybenko, Dorothy and Walter Gramm Professor of Engineering > > Dartmouth College, 8000 Cummings Hall, Hanover NH 03755 USA > > > > email: gvc at dartmouth.edu > > FAX: 603 646-9024 > > Cell: 603 369-1133 > > Assistant: Ellen Wirta, 603 646-9672, Ellen.K.Wirta at Dartmouth.edu > > > > -- > > > > > > > -- > > -- > > Hava T. Siegelmann, Ph.D. > > Provost Professor > > Director, BINDS Lab (Biologically Inspired Neural Dynamical Systems) > > Dept. of Computer Science > > Core member, Program of Neuroscience and Behavior > > University of Massachusetts Amherst > > Amherst, MA, 01003 > > Phone: 413-540-6826 > > LAB WEBSITE: http://binds.cs.umass.edu/ > > -------------- next part -------------- An HTML attachment was scrubbed... URL: From serafeim.perdikis at essex.ac.uk Wed Aug 26 10:45:10 2026 From: serafeim.perdikis at essex.ac.uk (Perdikis, Serafeim) Date: Wed, 26 Aug 2026 14:45:10 +0000 Subject: Connectionists: =?windows-1252?q?Essex_BCI-NE_webinar=3A_Designin?= =?windows-1252?q?g_optimal_training_protocols_and_feedback_by_Dr=2E_L=E9a?= =?windows-1252?q?_Pillette?= In-Reply-To: References: Message-ID: https://www.linkedin.com/company/essex-bcine-lab/ The Essex BCI-NE Lab invites you to join our next monthly webinar: Designing Neurofeedback: How to Train People to Control Their Own Brain Activity? Delivered by Dr. L?a Pillette SEAMLESS team, Research in Computer Science and Random Systems (IRISA), French National Centre for Scientific Research (CNRS), Rennes, France The webinar will take place over Zoom on Wednesday, 9th September 2026, at 2 PM UK time RSVP: https://www.linkedin.com/events/7498383525314789376 Abstract: Self-controlling one?s own brain activity is at the heart of neurofeedback and brain?computer interface (BCI) applications, such as supporting motor rehabilitation after stroke or controlling a videogame using only brain activity. Yet the mechanisms that enable people to learn this skill remain only partially understood. In this talk, I will present my previous research on how the feedback provided to users about their own brain activity shapes skill acquisition. Building on this, I will describe my current work investigating two main avenues to improve BCI user training. First, how to better exploit sensations, such as thermal ones, both to design innovative feedback, and to uncover new mental strategies for modulating brain activity. Second, how intelligent tutoring systems, i.e., digital technologies with educational aims, can be leveraged to personalize and improve BCI skill acquisition. Together, these lines of work highlight pathways toward more effective, individualized training protocols for brain self-regulation. Speaker Biography: L?a Pillette is a CNRS researcher and member of the Seamless team at IRISA, Rennes, France, since 2022. She obtained her PhD in computer science from the University of Bordeaux in 2019. Her research focuses on developing innovative methods to train individuals to regulate their brain activity, enabling more accessible and effective use of brain-computer interfaces for applications such as medical interventions and virtual world interactions. The Essex BCI-NE Lab webinars series takes place monthly (usually, on the second or third Wednesday of the month) over Zoom and is open to all. Speakers are invited to talk about their research for 45-50 minutes followed by Q&A. Where speakers allow it, we record the talks and make them available to everyone on our YouTube channel. You can watch previous talks at: https://www.youtube.com/@essexbcis If you don?t want to miss our next webinars, please email serafeim.perdikis at essex.ac.uk to ask to be added to our webinars mailing list. Best wishes, Simis Dr Serafeim Perdikis, Associate Professor (Senior Lecturer) Brain-Computer Interfaces and Neural Engineering Laboratory School of Computer Science and Electronic Engineering University of Essex Wivenhoe Park, Colchester CO4 3SQ United Kingdom -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: BCINEWebinarLeaPillete.png Type: image/png Size: 441064 bytes Desc: BCINEWebinarLeaPillete.png URL: From barak at pearlmutter.net Wed Aug 26 13:04:32 2026 From: barak at pearlmutter.net (Barak A. Pearlmutter) Date: Wed, 26 Aug 2026 18:04:32 +0100 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Dear Danny, If I might paraphrase that work, it seems to me the idea is that "subsymbolic" signals are used for local processing, while symbolic representations are used for longer distance communication, with things becoming more symbolic the longer and narrower the channel: from one brain region to another, from the brain to its future self, from one brain to another. That certainly seems consistent with the information bottleneck of Tishby et al, where distributions fragment into clustered representations as the bottleneck becomes more severe. My only issue is that, if you measure things in a conventional computer doing symbol processing, the individual bits on wires can be seen as subsymbolic. What is bit 7 of register 4? What is bit 11 of the address bus? It is not until you agglomerate things just right, at just the right level of abstraction, that the crisp symbolic nature of, say, a compiler running on a simple RISC, becomes apparent. This subproblem seems to subsume the whole question, because it's basically "figure out how it all works and then we can talk about whether there are symbolic representations." This might not obviate the distinction between symbolic and subsymbolic, but I'd argue that it does make it uninteresting. Cheers, --Barak. From robert.kentridge at durham.ac.uk Wed Aug 26 16:24:36 2026 From: robert.kentridge at durham.ac.uk (KENTRIDGE, ROBERT W.) Date: Wed, 26 Aug 2026 20:24:36 +0000 Subject: Connectionists: Connectionists In-Reply-To: References: Message-ID: Well, the Charlie Gross paper on Lettvin and the Grandmother cell expends many more lines (pages) on Konorski than on Lettvin. Konorski was an outstanding scientist. The 1934 Konorski & Miller paper on the associations underlying classical conditioning is probably the coolest paper I?ve ever read - it vies with Hecht, Schlaer, and Pirenne (1941?) as my favourite. cheers, Bob Sent from Outlook for Mac From: Gabriele Scheler Date: Wednesday, 26 August 2026 at 20:18 To: dst at cs.cmu.edu ; Michael Arbib ; Asim Roy ; KENTRIDGE, ROBERT W. ; solla at northwestern.edu Subject: Connectionists [EXTERNAL EMAIL] What happened to my post to connectionists? Awaiting moderation? And what is the joke about Marvin Minsky influencing scientology and calling himself the devil? Compared to that my post went straight to the topic: My published work addressing concept cells and distributed representations via high information nodes ("index cells"), my concern at suppressing researchers as exemplified by James Bowers' report, my suggestion at a more inclusive rather than narrow parochial view on neural network history and a valuable reference to Konorski as the originator of Hebbian learning - who was ostracized during Stalin's time, an experience that one should not want to repeat. Gabriele -- Dr. Gabriele Scheler Carl Correns Foundation for Mathematical Biology 1030 Judson Dr Mountain View, CA 94040 https://www.theoretical-biology.org Please re-send an email if I fail to respond. -------------- next part -------------- An HTML attachment was scrubbed... URL: From danny.silver at acadiau.ca Wed Aug 26 21:53:25 2026 From: danny.silver at acadiau.ca (Danny Silver) Date: Thu, 27 Aug 2026 01:53:25 +0000 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Dear Barak and Hello Stephen (great to speak with you again). I have attached my original email below for Stephen, as the last word from the connectionist mail list was that my original email was pending approval by the moderator. So not sure how Stephen saw your response but perhaps not my original email. Anyway .. the miracles of the web. Barak .. I like your paraphrase - you definitely get the major idea of the paper. Young children develop conceptual representations of things like mother, food they like, asking for more, or things that scare them - long before they have symbols for these concepts. But once they have learned such symbols (hand gestures or words) they can use them to communicate about these concepts to others, in albeit limited ways. Many lower-order animals seem to function in the world very well - learning about and using concepts - without ever having need for symbols. So symbols are not required to think, but they make explaining our thinking to others possible (kind of). But please note, there is a more subtle second idea that you may have missed. Animals (particularly humans) who use symbols to communicate externally with each other also gained the advantage of creating an additional constraint (a bias) for learning and reasoning about the world. As I say in the original email - what we learn and reason about is to some extent constrained/shaped by the ?lexicon? of symbols (and their related concepts) that we have already created or learned from others. Humans initially developed the ability to manipulate symbolic representations to communicate their thoughts to each other - it was necessary to survive. But then a beautiful thing happened, we started using the symbols and this language of thought to constrain/rationalize our thinking - or at least, at the best of times we do so {:]). Stephen, I can see how an ART network can develop conceptual feature representations, say for the concept from photos, before the network becomes aware of a symbol for , such as the spoken word ?cat". The associated recognition category may simply not have a label. But how is this conceptual representation later bound to a symbolic representation for the evert symbol ?cat?. This is particularly messy when one considers there are all kinds of cats - ones you can pet, one that can eat you. Perhaps through the simultaneity of both seeing a cat and also hear the word ?cat?. Hmmm ? Very interesting. ? Danny From: Grossberg, Stephen Date: Wednesday, August 26, 2026 at 3:58?PM To: Barak A. Pearlmutter , Danny Silver Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... CAUTION: This email comes from outside Acadia. Verify the sender and use caution with any requests, links or attachments. Evert recognition category provides a ? symbolic ? representation of the distributed feature pattern that it represents. A feature-category resonance links category and features into a bound state that enables conscious recognition of the features. My 2021 OUP book explains this in detail and provides lots of experimental support for it. Get Outlook for iOS ________________________________ From: Connectionists on behalf of Barak A. Pearlmutter Sent: Wednesday, 26 August 2026 13:04:32 To: Danny Silver Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Dear Danny, If I might paraphrase that work, it seems to me the idea is that "subsymbolic" signals are used for local processing, while symbolic representations are used for longer distance communication, with things becoming more symbolic the longer and narrower the channel: from one brain region to another, from the brain to its future self, from one brain to another. That certainly seems consistent with the information bottleneck of Tishby et al, where distributions fragment into clustered representations as the bottleneck becomes more severe. My only issue is that, if you measure things in a conventional computer doing symbol processing, the individual bits on wires can be seen as subsymbolic. What is bit 7 of register 4? What is bit 11 of the address bus? It is not until you agglomerate things just right, at just the right level of abstraction, that the crisp symbolic nature of, say, a compiler running on a simple RISC, becomes apparent. This subproblem seems to subsume the whole question, because it's basically "figure out how it all works and then we can talk about whether there are symbolic representations." This might not obviate the distinction between symbolic and subsymbolic, but I'd argue that it does make it uninteresting. Cheers, --Barak. -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: Re- Connectionists- AI at 50 videos....eml Type: application/octet-stream Size: 36330 bytes Desc: Re- Connectionists- AI at 50 videos....eml URL: From danny.silver at acadiau.ca Wed Aug 26 11:26:09 2026 From: danny.silver at acadiau.ca (Danny Silver) Date: Wed, 26 Aug 2026 15:26:09 +0000 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Dear Barrack .. I fully agree with your statements below, but would add to it the following: There is no question that the human nervous system has the ability to represent a simple symbol that is associated with a complex concept such as . We can use the written word ?cat' or spoken word "chat? or alternatively an iconic image of a cat's face as a symbol to convey a message about a , so our brains do learn representations for such. But this distributed representation came into being primarily for communication purposes forced by and helping to form the social nature of our existence with each other. Later in human development (as well as in several other animals), such symbolic representations (symreps) were found to be useful as an additional constraint on learning and reasoning about their more complex related concepts - which have a much more complex distributed representation (conreps) within the brain. In our 2023 paper (https://arxiv.org/abs/2304.13626) Tom Mitchell and I present a Neural-Symbolic Hypothesis: "Symbols are critical to intelligence NOT because they are the building blocks of thought, but because they are characterizations of thought that (1) allow us to explain our subsymbolic thinking to ourselves and others and (2) act as constraints on inference and learning about the world. Symbols explain our thinking and aid our thinking, but are not the foundation of our thinking." We conjecture that internal agent ?self-communication? using symreps meant for human-to-human communication, became key to human intelligence because: (1) it provides a second, more abstract level of representation and reasoning which can occur in parallel with subsymbolic reasoning, and (2) it places an additional constraint on learning where prior learning act as an inductive bias for learning new symbols and concepts. Shared symbols allow us to explain and justify, internally as well as externally, our decisions and actions. And what we learn is shaped and constrained by the ?lexicon? of what we recognize as symbols. For the readers digest version of the paper see the Neural-Symbolic Conference abstract at https://ceur-ws.org/Vol-3432/paper40.pdf Best regards ... Danny ========================== Daniel L. Silver PhD, CIM Professor Emeritus, Jodrey School of Computer Science Data Scientist, Acadia Institute for Data Analytics Acadia University, Office 314, Carnegie Hall, Wolfville, Nova Scotia Canada B4P 2R6 Cell: (902) 679-9315 acadiau.ca Facebook Twitter YouTube LinkedIn Flickr [id:image001.png at 01D366AF.7F868A70] From: Connectionists on behalf of Barak A. Pearlmutter Date: Tuesday, August 25, 2026 at 12:25?PM To: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... CAUTION: This email comes from outside Acadia. Verify the sender and use caution with any requests, links or attachments. > Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? > Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? > Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? People blithely toss out statements like that, but local uninterpretability does not follow from the definition of a distributed representation. It is very easy to construct distributed representations whose components *can* be understood in isolation (the bits of a standard binary representation of a natural number ?2??, say) or whose components cannot be understood in isolation (a code for a natural number ?2?? whose bits each have an associated set of half the numbers in the domain picked at random, say). The "big regions" distributed representation for points in 2D space from the PDP books is locally interpretable. Naturally, all other things being equal, given the choice we'd choose distributed representations whose parts are interpretable in isolation. Does the brain do that? Well, deep learning networks seem to, and experimental data seems consistent with that notion. --Barak Pearlmutter. -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image001.png Type: image/png Size: 7489 bytes Desc: image001.png URL: From steve at bu.edu Wed Aug 26 14:58:08 2026 From: steve at bu.edu (Grossberg, Stephen) Date: Wed, 26 Aug 2026 18:58:08 +0000 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Evert recognition category provides a ? symbolic ? representation of the distributed feature pattern that it represents. A feature-category resonance links category and features into a bound state that enables conscious recognition of the features. My 2021 OUP book explains this in detail and provides lots of experimental support for it. Get Outlook for iOS ________________________________ From: Connectionists on behalf of Barak A. Pearlmutter Sent: Wednesday, 26 August 2026 13:04:32 To: Danny Silver Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Dear Danny, If I might paraphrase that work, it seems to me the idea is that "subsymbolic" signals are used for local processing, while symbolic representations are used for longer distance communication, with things becoming more symbolic the longer and narrower the channel: from one brain region to another, from the brain to its future self, from one brain to another. That certainly seems consistent with the information bottleneck of Tishby et al, where distributions fragment into clustered representations as the bottleneck becomes more severe. My only issue is that, if you measure things in a conventional computer doing symbol processing, the individual bits on wires can be seen as subsymbolic. What is bit 7 of register 4? What is bit 11 of the address bus? It is not until you agglomerate things just right, at just the right level of abstraction, that the crisp symbolic nature of, say, a compiler running on a simple RISC, becomes apparent. This subproblem seems to subsume the whole question, because it's basically "figure out how it all works and then we can talk about whether there are symbolic representations." This might not obviate the distinction between symbolic and subsymbolic, but I'd argue that it does make it uninteresting. Cheers, --Barak. -------------- next part -------------- An HTML attachment was scrubbed... URL: From barak at pearlmutter.net Thu Aug 27 06:38:08 2026 From: barak at pearlmutter.net (Barak A. Pearlmutter) Date: Thu, 27 Aug 2026 11:38:08 +0100 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Dear Danny, > As I say in the original email - what we learn and reason about is to > some extent constrained/shaped by the ?lexicon? of symbols (and their > related concepts) that we have already created or learned from others. > Humans initially developed the ability to manipulate symbolic > representations to communicate their thoughts to each other - it was > necessary to survive. But then a beautiful thing happened, we started > using the symbols and this language of thought to constrain/rationalize our > thinking - or at least, at the best of times we do so {:]). This is coming from what I'd call the "introspection" strategy for figuring out how cognition works. How do we know the above? Through introspection. We can just feel those symbolic gears turning in our own heads. Bilingual subjects report switching which language they're thinking in. Etc. Despite its intuitive appeal, I'm not convinced that introspection is a reliable methodology for doing cognitive science. To give an example, even within what we classify as "symbolic" things like speech, people push continuous quantities onto the channel: volume, speed, cadence, mimicry of accents, deliberately slurred utterances, facial expressions, etc. We do this even in written forms. Handwriting is more expressive than typed. Even in typeset text, people use \emph{...} and \textbf{...} and manipulate size and \raisebox{...}{...}. It's as if there are highly intricate and complex underlying continuous quantities being compressed into a mostly-quantized channel. My working hypothesis would be that, even when symbolic things are happening in the brain, the discrete structures are engulfed in a cloud of associated probability distributions whose representation and manipulation dwarf, in computational terms, the tiny kernel of purely symbolic structure. Cheers, --Barak. On Thu, 27 Aug 2026 at 02:53, Danny Silver wrote: > Dear Barak and Hello Stephen (great to speak with you again). > I have attached my original email below for Stephen, as the last word from > the connectionist mail list was that my original email was pending > approval by the moderator. So not sure how Stephen saw your response but > perhaps not my original email. > Anyway .. the miracles of the web. > > Barak .. I like your paraphrase - you definitely get the major idea of the > paper. Young children develop conceptual representations of things like > mother, food they like, asking for more, or things that scare them - long > before they have symbols for these concepts. But once they have learned > such symbols (hand gestures or words) they can use them to communicate > about these concepts to others, in albeit limited ways. Many lower-order > animals seem to function in the world very well - learning about and using > concepts - without ever having need for symbols. So symbols are not > required to think, but they make explaining our thinking to others possible > (kind of). > > But please note, there is a more subtle second idea that you may have > missed. Animals (particularly humans) who use symbols to communicate > externally with each other also gained the *advantage of creating an > additional** constraint (a bias) for learning and reasoning about the > world*. As I say in the original email - what we learn and reason > about is to some extent constrained/shaped by the ?lexicon? of symbols (and > their related concepts) that we have already created or learned from > others. Humans initially developed the ability to manipulate symbolic > representations to communicate their thoughts to each other - it was > necessary to survive. But then a beautiful thing happened, we started > using the symbols and this language of thought to constrain/rationalize our > thinking - or at least, at the best of times we do so {:]). > > Stephen, I can see how an ART network can develop conceptual feature > representations, say for the concept from photos, before the network > becomes aware of a symbol for , such as the spoken word ?cat". The > associated recognition category may simply not have a label. But how is > this conceptual representation later bound to a symbolic representation for > the evert symbol ?cat?. This is particularly messy when one considers > there are all kinds of cats - ones you can pet, one that can eat you. > Perhaps through the simultaneity of both seeing a cat and also hear the > word ?cat?. Hmmm ? Very interesting. > > ? Danny > > *From: *Grossberg, Stephen > *Date: *Wednesday, August 26, 2026 at 3:58?PM > *To: *Barak A. Pearlmutter , Danny Silver < > danny.silver at acadiau.ca> > *Cc: *connectionists at mailman.srv.cs.cmu.edu < > connectionists at mailman.srv.cs.cmu.edu> > *Subject: *Re: Connectionists: AI at 50 videos... > > *CAUTION: *This email comes from outside Acadia. Verify the sender and > use caution with any requests, links or attachments. > Evert recognition category provides a ? symbolic ? representation of the > distributed feature pattern that it represents. A feature-category > resonance links category and features into a bound state that enables > conscious recognition of the features. My 2021 OUP book explains this in > detail and provides lots of experimental support for it. > > Get Outlook for iOS > ------------------------------ > *From:* Connectionists on > behalf of Barak A. Pearlmutter > *Sent:* Wednesday, 26 August 2026 13:04:32 > *To:* Danny Silver > *Cc:* connectionists at mailman.srv.cs.cmu.edu < > connectionists at mailman.srv.cs.cmu.edu> > *Subject:* Re: Connectionists: AI at 50 videos... > > Dear Danny, > > If I might paraphrase that work, it seems to me the idea is that > "subsymbolic" signals are used for local processing, while symbolic > representations are used for longer distance communication, with > things becoming more symbolic the longer and narrower the channel: > from one brain region to another, from the brain to its future self, > from one brain to another. > > That certainly seems consistent with the information bottleneck of > Tishby et al, where distributions fragment into clustered > representations as the bottleneck becomes more severe. > > My only issue is that, if you measure things in a conventional > computer doing symbol processing, the individual bits on wires can be > seen as subsymbolic. What is bit 7 of register 4? What is bit 11 of > the address bus? It is not until you agglomerate things just right, at > just the right level of abstraction, that the crisp symbolic nature > of, say, a compiler running on a simple RISC, becomes apparent. This > subproblem seems to subsume the whole question, because it's basically > "figure out how it all works and then we can talk about whether there > are symbolic representations." This might not obviate the distinction > between symbolic and subsymbolic, but I'd argue that it does make it > uninteresting. > > Cheers, > > --Barak. > -------------- next part -------------- An HTML attachment was scrubbed... URL: From danny.silver at acadiau.ca Thu Aug 27 07:24:27 2026 From: danny.silver at acadiau.ca (Danny Silver) Date: Thu, 27 Aug 2026 11:24:27 +0000 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Barack .. You are certainly correct that this is a theory based partially on introspection to fill in the gaps. However, you can see in the paper that there is some supportive evidence. And I most certainly agree with your last paragraph. .. Danny Sent from my Bell Samsung device over Canada?s largest network. ________________________________ From: Barak A. Pearlmutter Sent: Thursday, 27 August 2026 07:38:08 To: Danny Silver Cc: Grossberg, Stephen ; connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... CAUTION: This email comes from outside Acadia. Verify the sender and use caution with any requests, links or attachments. Dear Danny, As I say in the original email - what we learn and reason about is to some extent constrained/shaped by the ?lexicon? of symbols (and their related concepts) that we have already created or learned from others. Humans initially developed the ability to manipulate symbolic representations to communicate their thoughts to each other - it was necessary to survive. But then a beautiful thing happened, we started using the symbols and this language of thought to constrain/rationalize our thinking - or at least, at the best of times we do so {:]). This is coming from what I'd call the "introspection" strategy for figuring out how cognition works. How do we know the above? Through introspection. We can just feel those symbolic gears turning in our own heads. Bilingual subjects report switching which language they're thinking in. Etc. Despite its intuitive appeal, I'm not convinced that introspection is a reliable methodology for doing cognitive science. To give an example, even within what we classify as "symbolic" things like speech, people push continuous quantities onto the channel: volume, speed, cadence, mimicry of accents, deliberately slurred utterances, facial expressions, etc. We do this even in written forms. Handwriting is more expressive than typed. Even in typeset text, people use \emph{...} and \textbf{...} and manipulate size and \raisebox{...}{...}. It's as if there are highly intricate and complex underlying continuous quantities being compressed into a mostly-quantized channel. My working hypothesis would be that, even when symbolic things are happening in the brain, the discrete structures are engulfed in a cloud of associated probability distributions whose representation and manipulation dwarf, in computational terms, the tiny kernel of purely symbolic structure. Cheers, --Barak. On Thu, 27 Aug 2026 at 02:53, Danny Silver > wrote: Dear Barak and Hello Stephen (great to speak with you again). I have attached my original email below for Stephen, as the last word from the connectionist mail list was that my original email was pending approval by the moderator. So not sure how Stephen saw your response but perhaps not my original email. Anyway .. the miracles of the web. Barak .. I like your paraphrase - you definitely get the major idea of the paper. Young children develop conceptual representations of things like mother, food they like, asking for more, or things that scare them - long before they have symbols for these concepts. But once they have learned such symbols (hand gestures or words) they can use them to communicate about these concepts to others, in albeit limited ways. Many lower-order animals seem to function in the world very well - learning about and using concepts - without ever having need for symbols. So symbols are not required to think, but they make explaining our thinking to others possible (kind of). But please note, there is a more subtle second idea that you may have missed. Animals (particularly humans) who use symbols to communicate externally with each other also gained the advantage of creating an additional constraint (a bias) for learning and reasoning about the world. As I say in the original email - what we learn and reason about is to some extent constrained/shaped by the ?lexicon? of symbols (and their related concepts) that we have already created or learned from others. Humans initially developed the ability to manipulate symbolic representations to communicate their thoughts to each other - it was necessary to survive. But then a beautiful thing happened, we started using the symbols and this language of thought to constrain/rationalize our thinking - or at least, at the best of times we do so {:]). Stephen, I can see how an ART network can develop conceptual feature representations, say for the concept from photos, before the network becomes aware of a symbol for , such as the spoken word ?cat". The associated recognition category may simply not have a label. But how is this conceptual representation later bound to a symbolic representation for the evert symbol ?cat?. This is particularly messy when one considers there are all kinds of cats - ones you can pet, one that can eat you. Perhaps through the simultaneity of both seeing a cat and also hear the word ?cat?. Hmmm ? Very interesting. ? Danny From: Grossberg, Stephen > Date: Wednesday, August 26, 2026 at 3:58?PM To: Barak A. Pearlmutter >, Danny Silver > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: AI at 50 videos... CAUTION: This email comes from outside Acadia. Verify the sender and use caution with any requests, links or attachments. Evert recognition category provides a ? symbolic ? representation of the distributed feature pattern that it represents. A feature-category resonance links category and features into a bound state that enables conscious recognition of the features. My 2021 OUP book explains this in detail and provides lots of experimental support for it. Get Outlook for iOS ________________________________ From: Connectionists > on behalf of Barak A. Pearlmutter > Sent: Wednesday, 26 August 2026 13:04:32 To: Danny Silver > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: AI at 50 videos... Dear Danny, If I might paraphrase that work, it seems to me the idea is that "subsymbolic" signals are used for local processing, while symbolic representations are used for longer distance communication, with things becoming more symbolic the longer and narrower the channel: from one brain region to another, from the brain to its future self, from one brain to another. That certainly seems consistent with the information bottleneck of Tishby et al, where distributions fragment into clustered representations as the bottleneck becomes more severe. My only issue is that, if you measure things in a conventional computer doing symbol processing, the individual bits on wires can be seen as subsymbolic. What is bit 7 of register 4? What is bit 11 of the address bus? It is not until you agglomerate things just right, at just the right level of abstraction, that the crisp symbolic nature of, say, a compiler running on a simple RISC, becomes apparent. This subproblem seems to subsume the whole question, because it's basically "figure out how it all works and then we can talk about whether there are symbolic representations." This might not obviate the distinction between symbolic and subsymbolic, but I'd argue that it does make it uninteresting. Cheers, --Barak. -------------- next part -------------- An HTML attachment was scrubbed... URL: From jose at rubic.rutgers.edu Thu Aug 27 07:26:18 2026 From: jose at rubic.rutgers.edu (=?utf-8?B?U3RlcGhlbiBKb3PDqSBIYW5zb24=?=) Date: Thu, 27 Aug 2026 11:26:18 +0000 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> Asim, I sit here with my coffee neuron and my cup neuron creating a cup of coffee, and my table neuron holding the cup and coffee neurons representing a cup of coffee. On the table, on the floor, noting of course the cup of coffee is not on the floor but on the table. So this is a network of coffee, cups, tables, maybe coded by another neuron in a hierarchical network called morning coffee? Well this is the typical problem with localist model of the brain. How would it work exactly? So far just representing a single episodic event requires hierarchical network of neurons that generalize to what exactly? Do we need many such networks to represent all episodic memories of having a cup of coffee? This is similar to the problem with the socalled "fusiform face area" which with some proper controls doesn't really exist (cf Hanson 2022). What would a small blueberry are in fusiform gyrus be doing with all those faces we recognize? A brain face cloud server? Other accounts show that FG, LO, IN and PFC are at least involved in a network of areas (not neurons) in recognizing a face. But computationally what's the actual underlying functions? The shift to "neurons represent symbols' is not helpful to explain meaning. Remember, we have billions of neurons and 100 trillions connections between them. I'm not saying symbols aren't important and Harnad's Grounding arguments are becoming more critical (and need to be made more explicit) as LLMs start to be embedded in robots. Therefore an arguably simpler task, is to identify "symbols" in LLMs since we have no idea how they work either. Are symbols in their "neurons"? Cheers Stephen On 8/24/26 22:56, Asim Roy wrote: * Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? * Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? * Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195?223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -- [cid:part1.jnWCyG8r.5Hrjb0bP at rubic.rutgers.edu] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.jpg Type: image/jpeg Size: 57383 bytes Desc: signature.jpg URL: From jose at rubic.rutgers.edu Thu Aug 27 06:56:56 2026 From: jose at rubic.rutgers.edu (=?utf-8?B?U3RlcGhlbiBKb3PDqSBIYW5zb24=?=) Date: Thu, 27 Aug 2026 10:56:56 +0000 Subject: Connectionists: Connectionists In-Reply-To: References: Message-ID: Bob, Yes, this is a great paper. It was a critique of Skinner's "operant" reflex. Kornorski quite correctly pointed out that probably don't call it a reflex. Skinner had realized as compared to classical conditioning that operants were more of an equivalence class (generalization issues--and a backdoor to symbolic structures) . ie the referenced paper in BF Sknner, The generic nature of the concepts of stimulus and response. J. Gen. Psychol., 1935, 12, 40-65. Kornorski had also shown interactions between CC and OC. probably undermining Skinner's original claim he discovered OC. Skinner responded to Kornorski and Miller.. a very civilized debate. Stephen On 8/26/26 16:24, KENTRIDGE, ROBERT W. wrote: Well, the Charlie Gross paper on Lettvin and the Grandmother cell expends many more lines (pages) on Konorski than on Lettvin. Konorski was an outstanding scientist. The 1934 Konorski & Miller paper on the associations underlying classical conditioning is probably the coolest paper I?ve ever read - it vies with Hecht, Schlaer, and Pirenne (1941?) as my favourite. cheers, Bob Sent from Outlook for Mac From: Gabriele Scheler Date: Wednesday, 26 August 2026 at 20:18 To: dst at cs.cmu.edu ; Michael Arbib ; Asim Roy ; KENTRIDGE, ROBERT W. ; solla at northwestern.edu Subject: Connectionists [EXTERNAL EMAIL] What happened to my post to connectionists? Awaiting moderation? And what is the joke about Marvin Minsky influencing scientology and calling himself the devil? Compared to that my post went straight to the topic: My published work addressing concept cells and distributed representations via high information nodes ("index cells"), my concern at suppressing researchers as exemplified by James Bowers' report, my suggestion at a more inclusive rather than narrow parochial view on neural network history and a valuable reference to Konorski as the originator of Hebbian learning - who was ostracized during Stalin's time, an experience that one should not want to repeat. Gabriele -- Dr. Gabriele Scheler Carl Correns Foundation for Mathematical Biology 1030 Judson Dr Mountain View, CA 94040 https://www.theoretical-biology.org Please re-send an email if I fail to respond. -- [cid:part1.489rUfm9.DM4rmO8s at rubic.rutgers.edu] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.jpg Type: image/jpeg Size: 57383 bytes Desc: signature.jpg URL: From aidhwp1 at gmail.com Thu Aug 27 08:28:19 2026 From: aidhwp1 at gmail.com (aidhwp1) Date: Thu, 27 Aug 2026 21:28:19 +0900 Subject: Connectionists: IEEE HITC Awards - Call for Nomination - 2026 Message-ID: Dear Colleagues, Please accept our sincere apologies if you receive multiple copies of this message. The IEEE Technical Committee on Hyper-Intelligence (HITC) is a technical committee within IEEE Systems Council, aimed at fostering research and education in hyper-intelligent systems with applications. The IEEE Technical Committee on Hyper-Intelligence (HITC) solicits nominations for its 2026 awards. Each award includes a plaque presented at a selected IEEE HITC sponsored conference, along with a public citation on the IEEE HITC website. The committee solicits nominations of - Early Career Researcher (ECR) Award ( https://hyper-intelligence.org/award-early.html) - Middle Career Researcher (MCR) Award ( https://hyper-intelligence.org/award-middle.html) - Technical Achievement Award ( https://hyper-intelligence.org/award-achievement.html) - Outstanding PhD Dissertation Award ( https://hyper-intelligence.org/award-thesis.html) - Industrial Impact Award ( https://hyper-intelligence.org/award-industrial.html) ------------- Submissions: Nominations must be submitted via email (as a single PDF file) to the Award Selection Secretary, Xiaokang Zhou (zhou at kansai-u.ac.jp), and CC the Committee Chair, Laurence T. Yang (ltyang at ieee.org). The subject of the nomination email should be named as: HITC + Award Category + Nominee, e.g. HITC-Early Career Award-Obama. ------------- Important Dates: - Nomination Deadline: August 31, 2026 - Results Notification: September 21, 2026 ------------- Award Selection Committee: - Laurence T. Yang (Chair), Zhengzhou University, China (ltyang at ieee.org) - Flavia C. Delicato, Fluminense Federal University, Brazil ( felipe at cos.ufrj.br) - Bernady O. Apduhan, Kyushu Sangyo University, Japan (bob at is.kyusan-u.ac.jp ) - Anna Kobusi?ska, Poznan University of Technology, Poland ( Anna.Kobusinska at cs.put.poznan.pl) - Klimis Ntalianis, University of West Attica, Greece (kntal at image.ntua.gr) ------------- Award Selection Secretary: - Xiaokang Zhou, Kansai University, Japan (zhou at kansai-u.ac.jp) ------------- Award & Presentation Note: Awardees will be presented a plaque and will be recognized by IEEE HITC in its website, newsletter and archives. The awards for 2026 will be presented at CyberSciTech 2026 (IEEE HITC sponsored conference) - Melbourne, Australia, November 9-13, 2026. ============= IEEE HITC Award for Excellence for Early Career Researchers - 2026 The IEEE HITC Award for Excellence in Hyper-Intelligence (Early Career Researchers) recognizes up to 5 individuals who have made outstanding, influential, and potentially long-lasting contributions in the field of hyper-intelligent systems. Typically, the candidates are within 5 years of receiving their PhD degree as of January 01 of the year of the award. ------------- Nominations: A candidate may be nominated by members of the community. An individual may nominate at most one candidate for this award. A nomination application (as a single PDF file) should contain the following details: ? Name/email of person making the nomination (self-nominations are not eligible). ? Name/email of candidate for whom the award is recommended. ? A statement by the nominator (maximum of 500 words) as to why the nominee is highly deserving of the award both on excellence and in relation to IEEE HITC. ? CV of the nominee. ? Up to three support letters from persons other than the nominator ? these should be collected by the nominator and included in the nomination. Members of selection committee cannot be nominators or referees. ============= IEEE HITC Award for Excellence for Middle Career Researchers - 2026 The IEEE HITC Award for Excellence in Hyper-Intelligence (Middle Career Researcher) recognizes up to 3 individuals who have made distinguished, influential, and on-going yet towards long-lasting contributions in the field of hyper-intelligent systems with applications. Typically, the candidates are within 5 to 15 years of receiving their PhD degree as of January 01 of the year of the award. ------------- Nominations: A candidate must be nominated by members of the community. A nomination application (as a single PDF file) must consist of the following materials: ? Name/email of person making the nomination (self-nominations are not eligible). ? Name/email of the nominee. ? A statement by the nominator (maximum of 500 words) as to why the nominee is highly deserving of the award both on excellence and in relation to IEEE HITC. ? CV of the nominee. ? Up to three support letters from persons other than the nominator-these should be collected by the nominator and included in the nomination. ============= IEEE HITC Award for Excellence for Technical Achievement - 2026 The IEEE HITC Award for Excellence in Hyper-Intelligence (Technical Achievement) is awarded for significant and sustained contributions to the hyper-intelligent systems community through the Technical Committee on Hyper-Intelligence (HITC), coupled with an outstanding record of high quality and high impact research. The award contains a plaque. ------------- Nominations: A candidate may be nominated by colleagues/HITC members or may nominate him/her-self. An individual can nominate at most one candidate for this award. The candidate must be an IEEE and HITC member in good standing. A nomination application (as a single PDF file) should contain the following details: ? Professional Employment Affiliations: List the nominee's current professional affiliations and titles. ? Citation: Give a brief citation (thirty words or less) precisely stating the most salient reason(s) why the nominee is qualified for the award. ? Technical Contributions: Describe the nominee's technical achievements in scalable computing as well as significance and impact. (Max 2-page length) ? HITC Contribution: Describe the candidate?s service and specific contributions to HITC. (Max 2-page length) ? Endorsers: Each nomination must be supported by letter from three at least endorsers. An endorser can endorse only one candidate for this award. The endorsers will be required to comment on the nominee's technical contributions as well as service contributions to the HITC. The endorsement letters should be included in the nomination package. ============= IEEE HITC Outstanding PhD Dissertation Award - 2026 The IEEE HITC Outstanding PhD Dissertation Award is an annual award to recognize candidates that have recently received a PhD degree for no more than 2 years and have written an outstanding PhD dissertation in the field of the hyper-intelligent systems with applications. This award is established to encourage doctoral research that combines theory and practice or makes in-depth technical contributions, having the potential to contribute to the IEEE HITC. ------------- Nominations: A candidate may be nominated by members of the community. An individual may nominate at most one candidate for this award. A nomination application (as a single PDF file) must consist of the following materials: ? Name/email of person making the nomination (self-nominations are not eligible). ? Name/email of candidate. ? A statement by the nominator (maximum of 500 words) as to why the nominee is highly deserving of the award and a brief summary of the service contributions. ? CV of the nominee. ? A doctoral dissertation written by the applicant in any language, no more than 2 years prior to the submission deadline. ? A summary of the dissertation in English of up to 2 pages in length written by the PhD candidate, highlighting the significance of the problem, the technical approach taken, and the application context and potential. ? Sample published paper(s) in English based on the dissertation written primarily by the PhD candidate in scientific journals, especially IEEE Transactions/Journals. ? Listing of all publications by the applicant in the related field. ? A letter of recommendation from the applicant?s dissertation advisor that assesses the significance of the research, attests to the originality of the work, and comments on the engagement of the applicant in the HITC field. ? Up to three support letters from persons other than the nominator; these should be collected by the nominator and included in the nomination. ============= IEEE HITC Award for Excellence for Industrial Impact Award - 2026 The HITC Award for Excellence in Hyper-Intelligence (Industrial Impact Award) recognizes up to 3 individuals or small companies who have made distinguished, influential, and on-going yet towards long-lasting contributions in the field of hyper-intelligent systems with applications. ------------- Nominations: A candidate must be nominated by members of the community. A nomination application (as a single PDF file) must consist of the following materials: ? Name/email of person making the nomination (self-nominations are not eligible). ? Name/email of the nominee. ? A statement by the nominator (maximum of 500 words) as to why the nominee is highly deserving of the award both on excellence and in relation to IEEE HITC. ? CV of the nominee or supporting materials of the company showing the current impact/potential impact of the innovation (max 20-page length) ? Up to three support letters from persons other than the nominator-these should be collected by the nominator and included in the nomination. -------------- next part -------------- An HTML attachment was scrubbed... URL: From roman.bauer111 at gmail.com Thu Aug 27 09:15:22 2026 From: roman.bauer111 at gmail.com (roman bauer) Date: Thu, 27 Aug 2026 14:15:22 +0100 Subject: Connectionists: Academic positions in Artificial Intelligence at the University of Surrey In-Reply-To: References: Message-ID: Dear all, We are pleased to announce three open academic positions in Artificial Intelligence at the University of Surrey, UK: Professor in Artificial Intelligence (NICE Research Group within the Computer Science Research Centre) https://jobs.surrey.ac.uk/Vacancy.aspx?ref=037526 Senior Lecturer in Artificial Intelligence (NICE Research Group within the Computer Science Research Centre) https://jobs.surrey.ac.uk/Vacancy.aspx?ref=037426 Professor in Artificial Intelligence (Centre for Vision, Speech and Signal Processing) https://jobs.surrey.ac.uk/Vacancy.aspx?ref=037626 Closing date: Sunday 13 September 2026 The School of Computer Science and Electronic Engineering is seeking to recruit internationally recognised researchers to strengthen its research and teaching in Artificial Intelligence. These appointments form part of a wider recruitment campaign and strategic investment across the Faculty: https://www.jobs.ac.uk/enhanced/linking/university-of-surrey-engineering-physical-sciences-aug-2026/ The Professor and Senior Lecturer positions in Computer Science cover areas including Large Machine Learning Models, Neurosymbolic AI, Trustworthy Machine Learning, Natural Language Processing, and applications of AI and machine learning in automated reasoning, security, and software and systems development. Both posts are aligned with the Nature Inspired Computing and Engineering (NICE) Research Group within the Computer Science Research Centre. The Professor position within the Centre for Vision, Speech and Signal Processing focuses on leading growth in frontier AI, including multimodal foundation models, machine learning, agentic, people-centred and sustainable AI, with applications spanning human and animal health, biosciences, creative industries, sustainability, robotics and autonomous systems. The School is home to two established research centres with substantial expertise in AI and machine learning: the Computer Science Research Centre and the Centre for Vision, Speech and Signal Processing. Surrey has an established international reputation in AI research, ranking first in the UK for computer vision and among the top five for AI, computer vision, machine learning, robotics and natural language processing according to CSRankings.org. The School was also ranked seventh in the UK for Computer Science research outputs in REF2021. Computer Science and CVSSP are central to the Surrey Institute for People-Centred AI, a pan-University initiative bringing together AI research and expertise across health, engineering, social and behavioural sciences, business, law and the creative arts. The Institute leads a portfolio of more than ?100 million in grant awards, including major activities in healthcare and the creative industries, and two doctoral training programmes supporting more than 100 PhD researchers. Informal enquiries concerning the Computer Science positions may be directed to Professor Brijesh Dongol at b.dongol at surrey.ac.uk. Enquiries concerning the CVSSP position may be directed to Professor Adrian Hilton at a.hilton at surrey.ac.uk. Our staff and students come from around the world, and we are proud of our friendly and inclusive culture. The University is committed to building a diverse community, and applications from under-represented groups are particularly encouraged. Please share these opportunities with potentially interested candidates and across your networks. Best regards, Roman *Roman Bauer, Ph.D.* Senior Lecturer | Head of the NICE Research Group BioDynaMo Spokesperson (www.biodynamo.org) *A software suite to enable simulations of large, complex systems for research or policy making* *Computer Science Research Centre* University of Surrey | Guildford, UK r.bauer at surrey.ac.uk | ORCID | LinkedIn * Watch the **BioDynaMo introduction video* -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image001.jpg Type: image/jpeg Size: 60226 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image002.jpg Type: image/jpeg Size: 5608 bytes Desc: not available URL: From gscheler at gmail.com Thu Aug 27 08:39:04 2026 From: gscheler at gmail.com (Gabriele Scheler) Date: Thu, 27 Aug 2026 14:39:04 +0200 Subject: Connectionists: Symbols as indices into features representations Message-ID: https://pubmed.ncbi.nlm.nih.gov/40120001/ https://www.researchgate.net/publication/408161740_Non-standard_memory_models_with_indexed_retrieval -- Dr. Gabriele Scheler Carl Correns Foundation for Mathematical Biology 1030 Judson Dr Mountain View, CA 94040 https://www.theoretical-biology.org/#contribute Please re-send an email if I fail to respond. -------------- next part -------------- An HTML attachment was scrubbed... URL: From robert.kentridge at durham.ac.uk Thu Aug 27 09:30:17 2026 From: robert.kentridge at durham.ac.uk (KENTRIDGE, ROBERT W.) Date: Thu, 27 Aug 2026 13:30:17 +0000 Subject: Connectionists: AI@50 videos... In-Reply-To: <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> Message-ID: In Cavina-Pratesi et al (2010) (Cerebral Cortex 20:2319?2332) we suggest (along with others) that areas like FFA and those nearby are responding to particular combinations of texture, colour, and shape information, as our imagining results indicate. We found areas that responded selective only to these properties, but also plenty of areas (including FFA) that responded to combinations of them. Particular combinations might help in distinguishing between faces, but it looked to us as if these areas were still essentially sensory, not areas representing concepts. You might have to go a fair way further forward in the brain to find neurons that justify such claims. cheers Bob Sent from Outlook for Mac From: Connectionists on behalf of Stephen Jos? Hanson Date: Thursday, 27 August 2026 at 13:31 To: Asim Roy ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... [EXTERNAL EMAIL] Asim, I sit here with my coffee neuron and my cup neuron creating a cup of coffee, and my table neuron holding the cup and coffee neurons representing a cup of coffee. On the table, on the floor, noting of course the cup of coffee is not on the floor but on the table. So this is a network of coffee, cups, tables, maybe coded by another neuron in a hierarchical network called morning coffee? Well this is the typical problem with localist model of the brain. How would it work exactly? So far just representing a single episodic event requires hierarchical network of neurons that generalize to what exactly? Do we need many such networks to represent all episodic memories of having a cup of coffee? This is similar to the problem with the socalled "fusiform face area" which with some proper controls doesn't really exist (cf Hanson 2022). What would a small blueberry are in fusiform gyrus be doing with all those faces we recognize? A brain face cloud server? Other accounts show that FG, LO, IN and PFC are at least involved in a network of areas (not neurons) in recognizing a face. But computationally what's the actual underlying functions? The shift to "neurons represent symbols' is not helpful to explain meaning. Remember, we have billions of neurons and 100 trillions connections between them. I'm not saying symbols aren't important and Harnad's Grounding arguments are becoming more critical (and need to be made more explicit) as LLMs start to be embedded in robots. Therefore an arguably simpler task, is to identify "symbols" in LLMs since we have no idea how they work either. Are symbols in their "neurons"? Cheers Stephen On 8/24/26 22:56, Asim Roy wrote: * Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? * Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? * Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195?223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -- [cid:part1.jnWCyG8r.5Hrjb0bP at rubic.rutgers.edu] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.jpg Type: image/jpeg Size: 57383 bytes Desc: signature.jpg URL: From steve at bu.edu Thu Aug 27 11:00:19 2026 From: steve at bu.edu (Grossberg, Stephen) Date: Thu, 27 Aug 2026 15:00:19 +0000 Subject: Connectionists: AI@50 videos...: Almost 600 neural network modeling articles since 1957 model the main processes whereby our brains make our minds In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Dear Danny, It?s good to hear from you. You raise several issues. I reply to them in turn. In fact, ALL of the issues that you have described have been clarified over the years by biological neural networks that I and my colleagues have developed and used to provide principled explanations and quantitative computer simulations of large amounts of psychological and neurobiological data. You write below: "Young children develop conceptual representations of things like mother, food they like, asking for more, or things that scare them - long before they have symbols for these concepts. But once they have learned such symbols (hand gestures or words) they can use them to communicate about these concepts to others, in albeit limited ways." +++++++++++++++ You might find the following two articles relevant here: Grossberg, S. (2023). How children learn to understand language meanings: A neural model of adult?child multimodal interactions in real-time. Frontiers in Psychology, August 2, 2023. Section on Cognitive Science, Volume 14. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2023.1216479/full Grossberg, S. (2025). Neural network models of autonomous adaptive intelligence and artificial general intelligence: How our brains learn large language models and their meanings. Frontiers in Systems Neuroscience, July 29, 2025, Volume 19. https://www.frontiersin.org/journals/systems-neuroscience/articles/10.3389/fnsys.2025.1630151/full These articles clarify how children incrementally LEARN symbols and language utterances that have perceptual and affective MEANINGS by interacting with caregivers in real time. I also provide a self-contained and non-technical explanation of how this works in my 2026 book: YOUR CREATIVE BRAIN AND AI: How We Learn and Consciously Experience ART, MUSIC, and MEANING https://www.amazon.com/dp/0198965370?lv=shuf&channelId=500&plpRedirect=mhFallback +++++++++++++++++ You also wrote below: "Many lower-order animals seem to function in the world very well - learning about and using concepts - without ever having need for symbols. So symbols are not required to think, but they make explaining our thinking to others possible (kind of).? Primitive organisms like crustacea can function using stereotyped circuits, including what I have called context-sensitive avalanches; see Figure 4 in: Grossberg, S. (1970). Some networks that can learn, remember, and reproduce any number of complicated space-time patterns, II. Studies in Applied Mathematics, 49, 135-166. https://sites.bu.edu/steveg/files/2016/06/Gro1970SiAM.pdf or Chapter 5 in: Grossberg, S. (1974). Classical and instrumental learning by neural networks. In R. Rosen and F. Snell (Eds.), Progress in theoretical biology. New York: Academic Press, pp. 51-141. https://sites.bu.edu/steveg/files/2016/06/Gro1974ProgressTheorBiol.pdf Chapter 13 of my 2021 Magnum Opus CONSCIOUS MIND, RESONANT BRAIN: HOW EACH BRAIN MAKES A MIND https://www.amazon.com/dp/0190070552?lv=shuf&channelId=500&plpRedirect=mhFallback provides an evolutionary account of how avalanches have developed through phylogeny to replace stereotyped performance with learned sequential performance that is sensitive to multiple types of environmental feedback, leading to the cognitive and emotional circuits in human brains. These circuits do not, however, enable a primitive organism to ?think?. That requires brain circuits that support attention, learning, cognition, and action, which my other article model, including articles about Adaptive Resonance Theory. Another way that animals ?function in the world very well? is illustrated by terrestrial animals who can learn to navigate by using spatial maps and cognitive-emotional interactions. Many of these animals also use learned recognition categories to decide in what direction to navigate to achieve effective foraging. For example, Browning, A., Grossberg, S., and Mingolla, M. (2009). Cortical dynamics of navigation and steering in natural scenes: Motion-based object segmentation, heading, and obstacle avoidance. Neural Networks, 22, 1383-1398. https://sites.bu.edu/steveg/files/2016/06/BroGroMinNN2009.pdf Yet another article shows how prefrontal cortex can interact with multiple brain regions to enable humans and higher-animals to plan and act to realize valued goals at socially-appropriate times. For example, Grossberg, S. (2018). Desirability, availability, credit assignment, category learning, and attention: Cognitive-emotional and working memory dynamics of orbitofrontal, ventrolateral, and dorsolateral prefrontal cortices. Brain and Neuroscience Advances, May 8, 2018. https://journals.sagepub.com/doi/full/10.1177/2398212818772179 +++++++++++ You write below: "Stephen, I can see how an ART network can develop conceptual feature representations, say for the concept from photos, before the network becomes aware of a symbol for , such as the spoken word ?cat". The associated recognition category may simply not have a label. But how is this conceptual representation later bound to a symbolic representation for the evert symbol ?cat?. This is particularly messy when one considers there are all kinds of cats - ones you can pet, one that can eat you. "Perhaps through the simultaneity of both seeing a cat and also hear the word ?cat?. Hmmm ? Very interesting.? Danny, the above comments require several different interacting brain mechanisms to be explain and modeled. You write: "how an ART network can develop conceptual feature representations, say for the concept from photos, before the network becomes aware of a symbol for , such as the spoken word ?cat". The associated recognition category may simply not have a label.? The ARTMAP family of neural networks model how this works. They can incrementally learn in response to arbitrary sequences of unsupervised and supervised learning trials. For example: Carpenter, G.A., Grossberg, S., Markuzon, N., Reynolds, J.H., and Rosen, D.B. (1992). Fuzzy ARTMAP: A neural network architecture for incremental supervised learning of analog multidimensional maps. IEEE Transactions on Neural Networks, 3, 698-713. https://sites.bu.edu/steveg/files/2016/06/CarGroMarRey1992IEEETransNN.pdf You also write: ?there are all kinds of cats - ones you can pet, one that can eat you?. The process of vigilance control modulates ART category learning, with low vigilance leading to the learning of general, or abstract, recognition categories and high vigilance learning to the learning of specific, or concrete recognition categories, including exemplars. An automatic process called match tracking leads to minimax learning that conjointly maximizes category generality while minimizing predictive error. If vigilance is too low to predict the right answer when presented with one kind of cat, match tracking is triggered, thereby leading to hypothesis testing or memory search, for the other kind of cat. If the other kind of cat has not yet been learned, search ends an a novel category is learned to recognize the other kind of cat. Here is the first article that proved theorems for how vigilance control does this: Carpenter, G.A., and Grossberg, S. (1987). A massively parallel architecture for a self-organizing neural pattern recognition machine. Computer Vision, Graphics, and Image Processing, 37, 54-115. https://sites.bu.edu/steveg/files/2016/06/CarGro1987CVGIP.pdf This 1987 article predicted the existence of vigilance control. The following article shows that it exists, and explains psychological, anatomical, neurophysiological, biophysical, and biochemical data about how it works: Grossberg, S. and Versace, M. (2008). Spikes, synchrony, and attentive learning by laminar thalamocortical circuits. Brain Research, 1218, 278-312. https://sites.bu.edu/steveg/files/2016/06/GroVer2008BR.pdf See Figure 5. +++++++++++++++++++++ More generally, I have published almost 600 archival neural network modeling articles since 1957 that are downloadable from sites.bu.edu/steveg. They develop biological neural network models of the main processes whereby our brains make our conscious minds in health individuals and clinical patients. As I note on my web page, with over 100 gifted PhD students, postdocs, and faculty, I have discovered and led the development of "brain models of vision and visual object recognition; audition, speech, and language; development; attentive learning and memory; cognitive information processing and social cognition; reinforcement learning and motivation; cognitive-emotional interactions; navigation; sensory-motor control and robotics; and mental disorders. These models involve many parts of the brain, ranging from perception to action, and multiple levels of brain organization, ranging from individual spikes and their synchronization to cognition. Many of these projects are done in collaborations with PhD students, postdoctoral fellows, and faculty. I also collaborate with experimentalist colleagues to design experiments to test theoretical predictions and fill conceptually important gaps in the experimental literature, carry out analyses of the mathematical dynamics of neural systems, and transfer biological neural models to applications in neuromorphic engineering and technology.? I recommend to everyone that, before trying to develop a neural model of some brain process, you read the article titles and abstracts on sites.bu.edu/steveg. It would be a waste of your time and talent to just reinvent the wheel. It would be much more productive to build on the secure foundation of what is already known. Best, Steve Get Outlook for Mac From: Danny Silver Date: Wednesday, August 26, 2026 at 9:54?PM To: Grossberg, Stephen ; Barak A. Pearlmutter Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Dear Barak and Hello Stephen (great to speak with you again). I have attached my original email below for Stephen, as the last word from the connectionist mail list was that my original email was pending approval by the moderator. So not sure how Stephen saw your response but perhaps not my original email. Anyway .. the miracles of the web. Barak .. I like your paraphrase - you definitely get the major idea of the paper. Young children develop conceptual representations of things like mother, food they like, asking for more, or things that scare them - long before they have symbols for these concepts. But once they have learned such symbols (hand gestures or words) they can use them to communicate about these concepts to others, in albeit limited ways. Many lower-order animals seem to function in the world very well - learning about and using concepts - without ever having need for symbols. So symbols are not required to think, but they make explaining our thinking to others possible (kind of). But please note, there is a more subtle second idea that you may have missed. Animals (particularly humans) who use symbols to communicate externally with each other also gained the advantage of creating an additional constraint (a bias) for learning and reasoning about the world. As I say in the original email - what we learn and reason about is to some extent constrained/shaped by the ?lexicon? of symbols (and their related concepts) that we have already created or learned from others. Humans initially developed the ability to manipulate symbolic representations to communicate their thoughts to each other - it was necessary to survive. But then a beautiful thing happened, we started using the symbols and this language of thought to constrain/rationalize our thinking - or at least, at the best of times we do so {:]). Stephen, I can see how an ART network can develop conceptual feature representations, say for the concept from photos, before the network becomes aware of a symbol for , such as the spoken word ?cat". The associated recognition category may simply not have a label. But how is this conceptual representation later bound to a symbolic representation for the evert symbol ?cat?. This is particularly messy when one considers there are all kinds of cats - ones you can pet, one that can eat you. Perhaps through the simultaneity of both seeing a cat and also hear the word ?cat?. Hmmm ? Very interesting. ? Danny From: Grossberg, Stephen Date: Wednesday, August 26, 2026 at 3:58?PM To: Barak A. Pearlmutter , Danny Silver Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... CAUTION: This email comes from outside Acadia. Verify the sender and use caution with any requests, links or attachments. Evert recognition category provides a ? symbolic ? representation of the distributed feature pattern that it represents. A feature-category resonance links category and features into a bound state that enables conscious recognition of the features. My 2021 OUP book explains this in detail and provides lots of experimental support for it. Get Outlook for iOS ________________________________ From: Connectionists on behalf of Barak A. Pearlmutter Sent: Wednesday, 26 August 2026 13:04:32 To: Danny Silver Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Dear Danny, If I might paraphrase that work, it seems to me the idea is that "subsymbolic" signals are used for local processing, while symbolic representations are used for longer distance communication, with things becoming more symbolic the longer and narrower the channel: from one brain region to another, from the brain to its future self, from one brain to another. That certainly seems consistent with the information bottleneck of Tishby et al, where distributions fragment into clustered representations as the bottleneck becomes more severe. My only issue is that, if you measure things in a conventional computer doing symbol processing, the individual bits on wires can be seen as subsymbolic. What is bit 7 of register 4? What is bit 11 of the address bus? It is not until you agglomerate things just right, at just the right level of abstraction, that the crisp symbolic nature of, say, a compiler running on a simple RISC, becomes apparent. This subproblem seems to subsume the whole question, because it's basically "figure out how it all works and then we can talk about whether there are symbolic representations." This might not obviate the distinction between symbolic and subsymbolic, but I'd argue that it does make it uninteresting. Cheers, --Barak. -------------- next part -------------- An HTML attachment was scrubbed... URL: From caspar.schwiedrzik at googlemail.com Fri Aug 28 02:52:21 2026 From: caspar.schwiedrzik at googlemail.com (Caspar M. Schwiedrzik) Date: Fri, 28 Aug 2026 08:52:21 +0200 Subject: Connectionists: Deadline approaching: Postdoctoral researcher in Neuropixels recordings, predictive processing, and navigation In-Reply-To: References: Message-ID: We are looking for an outstanding postdoctoral researcher to establish and lead a new research program using Neuropixels recordings in behaving mice to study predictive processing at the interface of vision, space, and navigation. We will investigate how the brain learns regularities in structured environments and uses them to generate predictions about upcoming events. The successful candidate will develop experiments in which mice navigate controlled environments while encountering visual objects or events with learnable statistical structure. The goal is to understand how neural populations represent expectations, prediction errors, and the interaction between visual and spatial information during navigation. This new research program will be part of the lab?s broader cross-species research program on predictive processing and experience-dependent plasticity. Related experiments are conducted in the lab in non-human primates and humans, using electrophysiology, fMRI, (intracranial) EEG, and behavioral approaches. The position therefore offers the opportunity to develop mechanistic experiments in rodents while contributing to an integrated research program within the lab across species, methods, and spatial scales. The lab seeks to understand the cortical basis and computational principles of perception and experience-dependent plasticity in the brain. See https://doi.org/10.64898/2025.11.30.691163, https://doi.org/10.1038/s41467-024-51543-y, and https://doi.org/10.64898/2025.12.17.694881 for recent examples of our work. The postdoc will play a key role in expanding these efforts toward high-density electrophysiology in behaving mice. Experiments will take place at the new research building THINK ? Center for Theoretical and Integrative Neuroscience and Cognitive Science at Ruhr University Bochum (https://www.ruhr-uni-bochum.de/think/index.html.en). THINK provides a highly interdisciplinary environment for integrative neuroscience and cognitive science, with dedicated infrastructure for human and animal research, neuroimaging, behavioral experimentation, data analysis, and high-performance computing. The postdoc will join a growing team in the lab, including researchers working on electrophysiology, fMRI, and behavior in mice. The position is funded through the Research Center One Health Ruhr of the University Alliance Ruhr. One Health Ruhr brings together researchers from neuroscience, molecular biology, water research, cancer research, and related fields to study fundamental mechanisms of health and disease across biological and environmental scales. The new application deadline is 31.08.2026. Please find the official job ad and further information here: https://jobs.ruhr-uni-bochum.de/jobposting/a6646cd83e1d4ed7e6d72b4ead787ec28f531ce90 From a0091624 at gmail.com Thu Aug 27 20:49:43 2026 From: a0091624 at gmail.com (Mengmi Zhang) Date: Thu, 27 Aug 2026 20:49:43 -0400 Subject: Connectionists: call for NeurIPS DevAI workshop papers Message-ID: Hi all, We're organizing the first edition of *DevAI: Developmental Perspectives on AI*, a workshop at *NeurIPS 2026* in Atlanta on *12?13 December 2026*. The workshop brings together researchers from cognitive development, neuroscience, psychology, and AI to explore intelligence through a developmental lens: - What do human developmental trajectories reveal about intelligence? - How does current AI diverge from infant and child cognition? - How can developmental insights inform better AI systems? - How can AI serve as a tool for studying human development? We welcome papers, datasets, and benchmarks at the intersection of AI and human development. *Deadlines: **Sep 6, 2026* More information: DevAI Workshop We'd love to see a submission from you. Please also feel free to share this with students or colleagues who might be interested. -- Best, Mengmi Nanyang assistant professor Angela Goh Career Development Professor PI of Deep NeuroCognition Lab College of Computing and Data Science Nanyang Technological University (NTU), Singapore Lab website: https://a0091624.wixsite.com/deepneurocognition-1 Our lab: CIL@ N4-B1A-02; My office: N3-02c-89 @ NTU -------------- next part -------------- An HTML attachment was scrubbed... URL: From danny.silver at acadiau.ca Thu Aug 27 19:46:30 2026 From: danny.silver at acadiau.ca (Danny Silver) Date: Thu, 27 Aug 2026 23:46:30 +0000 Subject: Connectionists: AI@50 videos...: Almost 600 neural network modeling articles since 1957 model the main processes whereby our brains make our minds In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Dear Stephen. .. Thank you for taking the time to go through all of this and provide such detailed responses with references. It sounds like you agree that we learn both concepts and symbols that refer to concepts. And you are most certainly correct about building on the findings of others versus repeating such work. .. Danny Sent from my Bell Samsung device over Canada?s largest network. ________________________________ From: Grossberg, Stephen Sent: Thursday, 27 August 2026 12:00:19 To: Danny Silver ; Barak A. Pearlmutter Cc: connectionists at mailman.srv.cs.cmu.edu ; Grossberg, Stephen Subject: Re: Connectionists: AI at 50 videos...: Almost 600 neural network modeling articles since 1957 model the main processes whereby our brains make our minds CAUTION: This email comes from outside Acadia. Verify the sender and use caution with any requests, links or attachments. Dear Danny, It?s good to hear from you. You raise several issues. I reply to them in turn. In fact, ALL of the issues that you have described have been clarified over the years by biological neural networks that I and my colleagues have developed and used to provide principled explanations and quantitative computer simulations of large amounts of psychological and neurobiological data. You write below: "Young children develop conceptual representations of things like mother, food they like, asking for more, or things that scare them - long before they have symbols for these concepts. But once they have learned such symbols (hand gestures or words) they can use them to communicate about these concepts to others, in albeit limited ways." +++++++++++++++ You might find the following two articles relevant here: Grossberg, S. (2023). How children learn to understand language meanings: A neural model of adult?child multimodal interactions in real-time. Frontiers in Psychology, August 2, 2023. Section on Cognitive Science, Volume 14. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2023.1216479/full Grossberg, S. (2025). Neural network models of autonomous adaptive intelligence and artificial general intelligence: How our brains learn large language models and their meanings. Frontiers in Systems Neuroscience, July 29, 2025, Volume 19. https://www.frontiersin.org/journals/systems-neuroscience/articles/10.3389/fnsys.2025.1630151/full These articles clarify how children incrementally LEARN symbols and language utterances that have perceptual and affective MEANINGS by interacting with caregivers in real time. I also provide a self-contained and non-technical explanation of how this works in my 2026 book: YOUR CREATIVE BRAIN AND AI: How We Learn and Consciously Experience ART, MUSIC, and MEANING https://www.amazon.com/dp/0198965370?lv=shuf&channelId=500&plpRedirect=mhFallback +++++++++++++++++ You also wrote below: "Many lower-order animals seem to function in the world very well - learning about and using concepts - without ever having need for symbols. So symbols are not required to think, but they make explaining our thinking to others possible (kind of).? Primitive organisms like crustacea can function using stereotyped circuits, including what I have called context-sensitive avalanches; see Figure 4 in: Grossberg, S. (1970). Some networks that can learn, remember, and reproduce any number of complicated space-time patterns, II. Studies in Applied Mathematics, 49, 135-166. https://sites.bu.edu/steveg/files/2016/06/Gro1970SiAM.pdf or Chapter 5 in: Grossberg, S. (1974). Classical and instrumental learning by neural networks. In R. Rosen and F. Snell (Eds.), Progress in theoretical biology. New York: Academic Press, pp. 51-141. https://sites.bu.edu/steveg/files/2016/06/Gro1974ProgressTheorBiol.pdf Chapter 13 of my 2021 Magnum Opus CONSCIOUS MIND, RESONANT BRAIN: HOW EACH BRAIN MAKES A MIND https://www.amazon.com/dp/0190070552?lv=shuf&channelId=500&plpRedirect=mhFallback provides an evolutionary account of how avalanches have developed through phylogeny to replace stereotyped performance with learned sequential performance that is sensitive to multiple types of environmental feedback, leading to the cognitive and emotional circuits in human brains. These circuits do not, however, enable a primitive organism to ?think?. That requires brain circuits that support attention, learning, cognition, and action, which my other article model, including articles about Adaptive Resonance Theory. Another way that animals ?function in the world very well? is illustrated by terrestrial animals who can learn to navigate by using spatial maps and cognitive-emotional interactions. Many of these animals also use learned recognition categories to decide in what direction to navigate to achieve effective foraging. For example, Browning, A., Grossberg, S., and Mingolla, M. (2009). Cortical dynamics of navigation and steering in natural scenes: Motion-based object segmentation, heading, and obstacle avoidance. Neural Networks, 22, 1383-1398. https://sites.bu.edu/steveg/files/2016/06/BroGroMinNN2009.pdf Yet another article shows how prefrontal cortex can interact with multiple brain regions to enable humans and higher-animals to plan and act to realize valued goals at socially-appropriate times. For example, Grossberg, S. (2018). Desirability, availability, credit assignment, category learning, and attention: Cognitive-emotional and working memory dynamics of orbitofrontal, ventrolateral, and dorsolateral prefrontal cortices. Brain and Neuroscience Advances, May 8, 2018. https://journals.sagepub.com/doi/full/10.1177/2398212818772179 +++++++++++ You write below: "Stephen, I can see how an ART network can develop conceptual feature representations, say for the concept from photos, before the network becomes aware of a symbol for , such as the spoken word ?cat". The associated recognition category may simply not have a label. But how is this conceptual representation later bound to a symbolic representation for the evert symbol ?cat?. This is particularly messy when one considers there are all kinds of cats - ones you can pet, one that can eat you. "Perhaps through the simultaneity of both seeing a cat and also hear the word ?cat?. Hmmm ? Very interesting.? Danny, the above comments require several different interacting brain mechanisms to be explain and modeled. You write: "how an ART network can develop conceptual feature representations, say for the concept from photos, before the network becomes aware of a symbol for , such as the spoken word ?cat". The associated recognition category may simply not have a label.? The ARTMAP family of neural networks model how this works. They can incrementally learn in response to arbitrary sequences of unsupervised and supervised learning trials. For example: Carpenter, G.A., Grossberg, S., Markuzon, N., Reynolds, J.H., and Rosen, D.B. (1992). Fuzzy ARTMAP: A neural network architecture for incremental supervised learning of analog multidimensional maps. IEEE Transactions on Neural Networks, 3, 698-713. https://sites.bu.edu/steveg/files/2016/06/CarGroMarRey1992IEEETransNN.pdf You also write: ?there are all kinds of cats - ones you can pet, one that can eat you?. The process of vigilance control modulates ART category learning, with low vigilance leading to the learning of general, or abstract, recognition categories and high vigilance learning to the learning of specific, or concrete recognition categories, including exemplars. An automatic process called match tracking leads to minimax learning that conjointly maximizes category generality while minimizing predictive error. If vigilance is too low to predict the right answer when presented with one kind of cat, match tracking is triggered, thereby leading to hypothesis testing or memory search, for the other kind of cat. If the other kind of cat has not yet been learned, search ends an a novel category is learned to recognize the other kind of cat. Here is the first article that proved theorems for how vigilance control does this: Carpenter, G.A., and Grossberg, S. (1987). A massively parallel architecture for a self-organizing neural pattern recognition machine. Computer Vision, Graphics, and Image Processing, 37, 54-115. https://sites.bu.edu/steveg/files/2016/06/CarGro1987CVGIP.pdf This 1987 article predicted the existence of vigilance control. The following article shows that it exists, and explains psychological, anatomical, neurophysiological, biophysical, and biochemical data about how it works: Grossberg, S. and Versace, M. (2008). Spikes, synchrony, and attentive learning by laminar thalamocortical circuits. Brain Research, 1218, 278-312. https://sites.bu.edu/steveg/files/2016/06/GroVer2008BR.pdf See Figure 5. +++++++++++++++++++++ More generally, I have published almost 600 archival neural network modeling articles since 1957 that are downloadable from sites.bu.edu/steveg. They develop biological neural network models of the main processes whereby our brains make our conscious minds in health individuals and clinical patients. As I note on my web page, with over 100 gifted PhD students, postdocs, and faculty, I have discovered and led the development of "brain models of vision and visual object recognition; audition, speech, and language; development; attentive learning and memory; cognitive information processing and social cognition; reinforcement learning and motivation; cognitive-emotional interactions; navigation; sensory-motor control and robotics; and mental disorders. These models involve many parts of the brain, ranging from perception to action, and multiple levels of brain organization, ranging from individual spikes and their synchronization to cognition. Many of these projects are done in collaborations with PhD students, postdoctoral fellows, and faculty. I also collaborate with experimentalist colleagues to design experiments to test theoretical predictions and fill conceptually important gaps in the experimental literature, carry out analyses of the mathematical dynamics of neural systems, and transfer biological neural models to applications in neuromorphic engineering and technology.? I recommend to everyone that, before trying to develop a neural model of some brain process, you read the article titles and abstracts on sites.bu.edu/steveg. It would be a waste of your time and talent to just reinvent the wheel. It would be much more productive to build on the secure foundation of what is already known. Best, Steve Get Outlook for Mac From: Danny Silver Date: Wednesday, August 26, 2026 at 9:54?PM To: Grossberg, Stephen ; Barak A. Pearlmutter Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Dear Barak and Hello Stephen (great to speak with you again). I have attached my original email below for Stephen, as the last word from the connectionist mail list was that my original email was pending approval by the moderator. So not sure how Stephen saw your response but perhaps not my original email. Anyway .. the miracles of the web. Barak .. I like your paraphrase - you definitely get the major idea of the paper. Young children develop conceptual representations of things like mother, food they like, asking for more, or things that scare them - long before they have symbols for these concepts. But once they have learned such symbols (hand gestures or words) they can use them to communicate about these concepts to others, in albeit limited ways. Many lower-order animals seem to function in the world very well - learning about and using concepts - without ever having need for symbols. So symbols are not required to think, but they make explaining our thinking to others possible (kind of). But please note, there is a more subtle second idea that you may have missed. Animals (particularly humans) who use symbols to communicate externally with each other also gained the advantage of creating an additional constraint (a bias) for learning and reasoning about the world. As I say in the original email - what we learn and reason about is to some extent constrained/shaped by the ?lexicon? of symbols (and their related concepts) that we have already created or learned from others. Humans initially developed the ability to manipulate symbolic representations to communicate their thoughts to each other - it was necessary to survive. But then a beautiful thing happened, we started using the symbols and this language of thought to constrain/rationalize our thinking - or at least, at the best of times we do so {:]). Stephen, I can see how an ART network can develop conceptual feature representations, say for the concept from photos, before the network becomes aware of a symbol for , such as the spoken word ?cat". The associated recognition category may simply not have a label. But how is this conceptual representation later bound to a symbolic representation for the evert symbol ?cat?. This is particularly messy when one considers there are all kinds of cats - ones you can pet, one that can eat you. Perhaps through the simultaneity of both seeing a cat and also hear the word ?cat?. Hmmm ? Very interesting. ? Danny From: Grossberg, Stephen Date: Wednesday, August 26, 2026 at 3:58?PM To: Barak A. Pearlmutter , Danny Silver Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... CAUTION: This email comes from outside Acadia. Verify the sender and use caution with any requests, links or attachments. Evert recognition category provides a ? symbolic ? representation of the distributed feature pattern that it represents. A feature-category resonance links category and features into a bound state that enables conscious recognition of the features. My 2021 OUP book explains this in detail and provides lots of experimental support for it. Get Outlook for iOS ________________________________ From: Connectionists on behalf of Barak A. Pearlmutter Sent: Wednesday, 26 August 2026 13:04:32 To: Danny Silver Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Dear Danny, If I might paraphrase that work, it seems to me the idea is that "subsymbolic" signals are used for local processing, while symbolic representations are used for longer distance communication, with things becoming more symbolic the longer and narrower the channel: from one brain region to another, from the brain to its future self, from one brain to another. That certainly seems consistent with the information bottleneck of Tishby et al, where distributions fragment into clustered representations as the bottleneck becomes more severe. My only issue is that, if you measure things in a conventional computer doing symbol processing, the individual bits on wires can be seen as subsymbolic. What is bit 7 of register 4? What is bit 11 of the address bus? It is not until you agglomerate things just right, at just the right level of abstraction, that the crisp symbolic nature of, say, a compiler running on a simple RISC, becomes apparent. This subproblem seems to subsume the whole question, because it's basically "figure out how it all works and then we can talk about whether there are symbolic representations." This might not obviate the distinction between symbolic and subsymbolic, but I'd argue that it does make it uninteresting. Cheers, --Barak. -------------- next part -------------- An HTML attachment was scrubbed... URL: From swickbrennan at yahoo.com Thu Aug 27 11:38:14 2026 From: swickbrennan at yahoo.com (Brennan Swick) Date: Thu, 27 Aug 2026 15:38:14 +0000 (UTC) Subject: Connectionists: [CFP] Final Call for Papers - NeuRo-SymBolic World Models (RoBoWoMo) @ IROS 2026 References: <1597868863.1622905.1787845094310.ref@mail.yahoo.com> Message-ID: <1597868863.1622905.1787845094310@mail.yahoo.com> TL;DR: Event: NeuRo-SymBolic World Models (RoBoWoMo) Workshop at IROS 2026 (Pittsburgh, US) Date: September 27, 2026 Website: https://worldmodelworkshop.github.io/ Submission Deadline: August 25, 2026 September 1, 2026 OpenReview submission link: https://bit.ly/SubmitToRoBoWoMo? We are organizing the NeuRo-SymBolic World Models (RoBoWoMo) Workshop at IROS 2026 (Sept 27 in Pittsburgh) and would love for you to submit! Call for Papers:? - Half Papers: Up to 4 pages (excluding references/appendices) - Full Papers: Up to 8 pages (excluding references/appendices) We welcome submissions covering neuro-symbolic unification, hybrid architectures, task-driven world modeling, benchmarking/evaluation, etc. All accepted papers will have the opportunity to be presented via lightning talks and a dedicated poster session, with the strongest submissions selected for longer spotlight presentations. For more details on formatting, submission policies, and topics of interest, please visit our Call for Papers page: https://worldmodelworkshop.github.io/call_for_papers/? ? Important Dates (AoE): - Submission Deadline: August 25, 2026 September 1, 2026 - Author Notification: September 10, 2026 - Camera-Ready Version Deadline: September 21, 2026 - Workshop Date: September 27, 2026 For schedule details: https://worldmodelworkshop.github.io/schedule? ? Invited Speakers and Panelists:? - Sherry Yang (New York University & Google DeepMind) - Yilun Du (Harvard University) - Tom Silver & Yixuan Huang (Princeton University) - Siddharth Srivastava (Arizona State University) - Emre Ugur (Bogazici University) - Sungjin Ahn (KAIST & New York University) - Jiajun Wu (Stanford University) -------------- next part -------------- An HTML attachment was scrubbed... URL: From ASIM.ROY at asu.edu Thu Aug 27 16:47:04 2026 From: ASIM.ROY at asu.edu (Asim Roy) Date: Thu, 27 Aug 2026 20:47:04 +0000 Subject: Connectionists: AI@50 videos... In-Reply-To: <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> Message-ID: Stephen, You raise important issues. There are many ways of dealing with them. Here?s one of the ways. The image below shows what DARPA wanted for Explainable AI. Essentially, find the parts of an object for its prediction. So, for a cat, verify some of its unique parts like the fur, whiskers, and claws. Thus, in our explainable models, we ?teach? an object detection model about these parts. What we get as outputs, for a cat, are predictions for the cat and its parts. There?s a layer of logic that uses these outputs and verifies the existence of the parts before ?finally? predicting that there?s a cat in the image. In essence, it?s neuro-symbolic. One can do the symbolic part with a graphical model too. We can also predict that there?s a cat even if we don?t ?see? all of the parts, such as when we just see its face and not the rest of the body. And, in a similar way, your ?morning coffee? concept can be built from its parts that you mention. There?s a long tradition in computer vision of finding parts of objects inside the model. Our initial conception was that the higher-level filters in a CNN correspond to certain abstractions such as a nose, eyes, ears for a human. That was the distributed representation but based on lower-level abstractions (parts of objects). But years of research showed that one could not find these kinds of lower-level abstractions in standard CNN models. So, what do you do? Well, one way is to force some of the filters to correspond to these parts. There?s a long history to this body of work in computer vision, including that of Hinton, to create abstractions within a model. Both ways, we are ?teaching? the models about ?low-level abstraction.? In a way, in both ways, it?s a hierarchical system. In both ways, we create a neuro-symbolic system. By the way, abstractions are generalizations, they don?t code specific episodic memories. Your ?table,? ?coffee,? ?cup? and all that are abstract notions. Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [A screenshot of a computer AI-generated content may be incorrect.] From: Stephen Jos? Hanson Sent: Thursday, August 27, 2026 4:26 AM To: Asim Roy ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Asim, I sit here with my coffee neuron and my cup neuron creating a cup of coffee, and my table neuron holding the cup and coffee neurons representing a cup of coffee. On the table, on the floor, noting of course the cup of coffee is not on the floor but on the table. So this is a network of coffee, cups, tables, maybe coded by another neuron in a hierarchical network called morning coffee? Well this is the typical problem with localist model of the brain. How would it work exactly? So far just representing a single episodic event requires hierarchical network of neurons that generalize to what exactly? Do we need many such networks to represent all episodic memories of having a cup of coffee? This is similar to the problem with the socalled "fusiform face area" which with some proper controls doesn't really exist (cf Hanson 2022). What would a small blueberry are in fusiform gyrus be doing with all those faces we recognize? A brain face cloud server? Other accounts show that FG, LO, IN and PFC are at least involved in a network of areas (not neurons) in recognizing a face. But computationally what's the actual underlying functions? The shift to "neurons represent symbols' is not helpful to explain meaning. Remember, we have billions of neurons and 100 trillions connections between them. I'm not saying symbols aren't important and Harnad's Grounding arguments are becoming more critical (and need to be made more explicit) as LLMs start to be embedded in robots. Therefore an arguably simpler task, is to identify "symbols" in LLMs since we have no idea how they work either. Are symbols in their "neurons"? Cheers Stephen On 8/24/26 22:56, Asim Roy wrote: 1. Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? 2. Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? 3. Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195?223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -- [cid:image003.jpg at 01DD3625.F5884020] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image003.jpg Type: image/jpeg Size: 57383 bytes Desc: image003.jpg URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image004.jpg Type: image/jpeg Size: 45946 bytes Desc: image004.jpg URL: From giovanni.stanco at unina.it Thu Aug 27 17:21:09 2026 From: giovanni.stanco at unina.it (GIOVANNI STANCO) Date: Thu, 27 Aug 2026 21:21:09 +0000 Subject: Connectionists: [CFP: IWNC'26 - CNSM'26] Extended Deadline - Third International Workshop on Integrated Wireless Networking and Computing 2026 In-Reply-To: References: Message-ID: The submission deadline for workshop papers is August 31, 2026 (Extended). Third International Workshop on Integrated Wireless Networking and Computing 2026 (IWNC'26) Joint with the 22nd International Conference on Network and Service Management (CNSM 2026) Alcal? de Henares (Madrid), Spain // 26 - 30 October, 2026 https://sites.google.com/view/iwnc-2026/ CALL FOR PAPERS The Third International Workshop on Integrated Wireless Networking and Computing 2026 (IWNC'26) invites high-quality submissions of papers describing original and unpublished research results regarding the use of intelligent computing capabilities at the edge or within the network. TOPICS OF INTEREST Topics of interest include, but are not limited to: ? AI/ML-driven optimization in wireless systems ? Management and orchestration architectures and techniques for next-gen networks ? Orchestration solutions for control plane, core, and RAN ? Joint design of communication and computation functions ? Programmable wireless networks and smart NICs ? In-Network Computing architectures and applications ? Cross-layer and cross-domain design approaches ? Early-stage research ideas and system prototypes on Integrated Wireless Networking and Computing ? Testbeds, emulators, and benchmarking for wireless networks ? Wireless networks for AI applications ? AI and ML in Edge Computing ? Advancements in wireless networks for Federated Learning ? Resilience, scalability, and adaptability of next-gen networks ? Cybersecurity Challenges in Integrated Wireless Networking and Computing ? Security and Privacy in Wireless Networks ? Energy-aware and Sustainable Wireless Networks ? O-RAN and MEC solutions ? O-RAN orchestration and management SUBMISSION GUIDELINES Submitted manuscripts should use IEEE 2-column conference style and are limited to 6 pages (including references). All papers accepted by a workshop will be published in the CNSM 2026 Proceedings and will be sent for inclusion in the IEEE eXplore digital library. The submission and revision process will be managed through the EDAS system (submission link: https://edas.info/N35574). IMPORTANT DATES ? Workshop papers submission deadline: August 31, 2026 ? Notification of acceptance: September 11, 2026 ? Camera-ready deadline: September 18, 2026 ? Workshop: October 30, 2026 We look forward to your valuable contributions and hope to see you at the workshop. Best regards, Stefania Zinno and Giovanni Stanco Third International Workshop on Integrated Wireless Networking and Computing 2026 (IWNC'26) -------------- next part -------------- An HTML attachment was scrubbed... URL: From agaster at pme.duth.gr Fri Aug 28 04:00:48 2026 From: agaster at pme.duth.gr (Antonios Gasteratos) Date: Fri, 28 Aug 2026 11:00:48 +0300 Subject: Connectionists: =?utf-8?q?ECMR_2027_=7C_Thessaloniki=2C_Greece_?= =?utf-8?q?=7C_31_Aug=E2=80=933_Sept_2027?= Message-ID: Dear colleagues, We are pleased to announce that the European Conference on Mobile Robots (ECMR 2027) will take place in Thessaloniki, Greece, from 31 August to 3 September 2027. Please save the date and feel free to share the announcement with colleagues who may be interested. Further information, including the Call for Papers and important dates, will follow soon. We look forward to welcoming the mobile robotics community to Thessaloniki! ECMR 2027 31 August ? 3 September 2027 Thessaloniki, Greece www.ecmr2027.gr Best regards, Antonios Gasteratos, General Chair Giannis Kostavelis, General Co-Chair Kostas Alexis, Program Chair ? -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ECRM2027 SAVE THE DATE-01.jpeg Type: image/jpeg Size: 324573 bytes Desc: not available URL: From emergingtechnetwork.publicity at gmail.com Fri Aug 28 07:59:29 2026 From: emergingtechnetwork.publicity at gmail.com (=?UTF-8?Q?Gizem_G=C3=BCltekin_Varkonyi?=) Date: Fri, 28 Aug 2026 14:59:29 +0300 Subject: Connectionists: SNAMS 2026: The IEEE co-sponsored 13th Conference on Social Networks Analysis, Management and Security, 17 - 20 November 2026 | Barcelona, Spain Message-ID: [Apologies if you got multiple copies of this invitation] The 13th International Conference on Social Networks Analysis, Management and Security(SNAMS-2026) https://emergingtechnet.org/SNAMS2026/index.php 17 - 20 November 2026 | Barcelona, Spain (Technically Co-Sponsored by IEEE Spain Section) *SNAMS 2026 CFP* Social network analysis is concerned with the study of relationships between social entities. The recent advances in internet technologies and social media sites, such as Facebook, Twitter and LinkedIn, have created outstanding opportunities for individuals to connect, communicate or comment on issues or events of their interests. Social networks are dynamic and evolving in nature; they also involve a huge number of users. Frequently, the information related to a certain concept is distributed among several servers. This brings numerous challenges to researchers, particularly in the data mining and machine learning fields. SNAMS 2026 aims to investigate the opportunities and in all aspects of Social Networks. In addition, it seeks for novel contributions that help mitigate SNAMS challenges. That is, the objective of SNAMS 2026 is to provide the opportunity for students, scientists, engineers, and researchers to discuss and exchange new ideas, novel results and experience on all aspects of Social Networks. Researchers are encouraged to submit original research contributions in all major areas, which include, but not limited to: *SYSTEMS & INFRASTRUCTURE* Systems and algorithms for social search Infrastructure support for social networks and systems Dynamics and evolution patterns of large and complex networks Social properties in systems design Learnings from operational social networks Big Data and Social Paradigms *ALGORITHMS AND MODELS* Deep Learning and Knowledge Discovery. Measurement and analysis of social and crowdsourcing systems Benchmarking, modeling, performance and workload characterization Modeling Social Networks and behavior Management of social network data Methods for social and media analysis Information propagation and assimilation in social networks Data mining and machine learning in social systems *APPLICATIONS* Novel social applications and systems Transient OSNs (e.g. Snapchat) Special purpose OSNs (e.g., Instagram, Vine) Communities in social networks Collaboration networks New models of advertising and monetization in social networks *PRIVACY & SECURITY* Privacy and security in social systems Trust and reputations in social systems Detection, analysis, prevention of spam, phishing, and misbehavior in social systems * Submission Types:* Accepted types of submissions are including: * Full papers (8 Pages), * Short papers and PhD forum (6 Pages) * Workshop papers (6 Pages) * Posters and demos (2 Pages). All accepted papers will be published in the conference proceedings and submitted to IEEE for publication and to be indexed by Scopus. *SNAMS 2026 Joint Workshops * The 14th workshop on Big Data and Social Networking Management and Security (BDSN 2026) The 10th Workshop on Data Science Engineering and its Applications (DSEA 2026) The 12th Workshop on Online Social Networks Technologies (OSNT 2026) The 10th Workshop on Advances in Natural Language Processing (ANLP 2026) The 9th Workshop on Sentiment Analysis and Mining of Social Networks (SAMSN 2026) The 4th Workshop on Cognitive and Neural Systems (CNS 2026) *Submissions Guidelines and Proceedings* Manuscripts should be prepared in 10-point font using the IEEE 8.5" x 11" two-column format (IEEE Templates ). All papers should be in PDF format, and submitted electronically at Paper Submission Link. A full paper must not exceed the stated length (including all figures, tables and references). Submitted papers must present original unpublished research that is not currently under review for any other conference or journal. Authors may contact the Program Chair for further information or clarification. All submissions are peer-reviewed by at least three reviewers. Accepted papers will appear in the SNAMS Proceedings, and be submitted to IEEE for inclusion. *Important Dates* ? *Paper submission deadline: * *September 10, 2026 * *(Firm and Final) * ? Notification to Authors: 25 September 2026 ? Camera Ready and Registration: 10 October 2026 Please send any inquiry to the SNAMS Team at: emergingtechnetwork at gmail.com -------------- next part -------------- An HTML attachment was scrubbed... URL: From steve at bu.edu Fri Aug 28 11:00:39 2026 From: steve at bu.edu (Grossberg, Stephen) Date: Fri, 28 Aug 2026 15:00:39 +0000 Subject: Connectionists: AI@50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> Message-ID: Dear Asim, Thanks for bringing up the DARPA Explainable AI program. I was invited to submit an article to a Special Issue about Explainable AI that was published when the DARPA program was announced. The editors of the Special Issue knew that Adaptive Resonance Theory, or ART, models automatically discover and incrementally learn over multiple learning trials to pay attention to the predictive, or explainable, critical feature patterns that control successful predictions and are used in learning by the adaptive weights in bottom-up adaptive filters and top-down expectations. ART does this AUTOMATICALLY. You seem to have to use an external teacher to do it. Here is the article: Grossberg, S. (2020). A path towards Explainable AI and autonomous adaptive intelligence: Deep Learning, Adaptive Resonance, and models of perception, emotion, and action. Frontiers in Neurobotics, June 25, 2020. https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2020.00036/full The article?s Abstract illustrates its explanatory range [boldface mine]: * "Biological neural network models whereby brains make minds help to understand autonomous adaptive intelligence. This article summarizes why the dynamics and emergent properties of such models for perception, cognition, emotion, and action are explainable, and thus amenable to being confidently implemented in large-scale applications. Key to their explainability is how these models combine fast activations, or short-term memory (STM) traces, and learned weights, or long-term memory (LTM) traces. Visual and auditory perceptual models have explainable conscious STM representations of visual surfaces and auditory streams in surface-shroud resonances and stream-shroud resonances, respectively. Deep Learning is often used to classify data. However, Deep Learning can experience catastrophic forgetting: At any stage of learning, an unpredictable part of its memory can collapse. Even if it makes some accurate classifications, they are not explainable and thus cannot be used with confidence. Deep Learning shares these problems with the back propagation algorithm, whose computational problems due to non-local weight transport during mismatch learning were described in the 1980s. Deep Learning became popular after very fast computers and huge online databases became available that enabled new applications despite these problems. Adaptive Resonance Theory, or ART, algorithms overcome the computational problems of back propagation and Deep Learning. ART is a self-organizing production system that incrementally learns, using arbitrary combinations of unsupervised and supervised learning and only locally computable quantities, to rapidly classify large non-stationary databases without experiencing catastrophic forgetting. ART classifications and predictions are explainable using the attended critical feature patterns in STM on which they build. The LTM adaptive weights of the fuzzy ARTMAP algorithm induce fuzzy IF-THEN rules that explain what feature combinations predict successful outcomes. ART has been successfully used in multiple large-scale real-world applications, including remote sensing, medical database prediction, and social media data clustering. Also explainable are the MOTIVATOR model of reinforcement learning and cognitive-emotional interactions, and the VITE, DIRECT, DIVA, and SOVEREIGN models for reaching, speech production, spatial navigation, and autonomous adaptive intelligence. These biological models exemplify complementary computing, and use local laws for match learning and mismatch learning that avoid the problems of Deep Learning." Chapter 5 in my 2021 Magnum Opus CONSCIOUS MIND, RESONANT BRAIN: HOW EACH BRAIN MAKES A MIND https://www.amazon.com/dp/0190070552?lv=shuf&channelId=500&plpRedirect=mhFallback provides a self-contained and non-technical overview and synthesis of my results about ART and Explainable AI. The other chapters summarize neural network models of the main processes whereby our brains make our conscious minds in healthy individuals and clinical patients. Best, Steve Stephen Grossberg Wang Professor of Cognitive and Neural Systems Director, Center for Adaptive Systems Emeritus Professor of Mathematics & Statistics, Psychological & Brain Sciences, and Biomedical Engineering Boston University sites.bu.edu/steveg/ steve at bu.edu http://en.wikipedia.org/wiki/Stephen_Grossberg http://scholar.google.com/citations?user=3BIV70wAAAAJ&hl=en https://sites.bu.edu/steveg/files/2021/08/Grossberg-CV-8-14-21.pdf https://youtu.be/9n5AnvFur7I https://www.youtube.com/watch?v=_hBye6JQCh4 https://www.amazon.com/Conscious-Mind-Resonant-Brain-Makes/dp/0190070552 https://www.amazon.com/Your-Creative-Brain-Consciously-Experience/dp/0198965370 From: Connectionists on behalf of Asim Roy Date: Friday, August 28, 2026 at 3:51?AM To: Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Stephen, You raise important issues. There are many ways of dealing with them. Here?s one of the ways. The image below shows what DARPA wanted for Explainable AI. Essentially, find the parts of an object for its prediction. So, for a cat, verify some of its unique parts like the fur, whiskers, and claws. Thus, in our explainable models, we ?teach? an object detection model about these parts. What we get as outputs, for a cat, are predictions for the cat and its parts. There?s a layer of logic that uses these outputs and verifies the existence of the parts before ?finally? predicting that there?s a cat in the image. In essence, it?s neuro-symbolic. One can do the symbolic part with a graphical model too. We can also predict that there?s a cat even if we don?t ?see? all of the parts, such as when we just see its face and not the rest of the body. And, in a similar way, your ?morning coffee? concept can be built from its parts that you mention. There?s a long tradition in computer vision of finding parts of objects inside the model. Our initial conception was that the higher-level filters in a CNN correspond to certain abstractions such as a nose, eyes, ears for a human. That was the distributed representation but based on lower-level abstractions (parts of objects). But years of research showed that one could not find these kinds of lower-level abstractions in standard CNN models. So, what do you do? Well, one way is to force some of the filters to correspond to these parts. There?s a long history to this body of work in computer vision, including that of Hinton, to create abstractions within a model. Both ways, we are ?teaching? the models about ?low-level abstraction.? In a way, in both ways, it?s a hierarchical system. In both ways, we create a neuro-symbolic system. By the way, abstractions are generalizations, they don?t code specific episodic memories. Your ?table,? ?coffee,? ?cup? and all that are abstract notions. Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [A screenshot of a computer AI-generated content may be incorrect.] From: Stephen Jos? Hanson Sent: Thursday, August 27, 2026 4:26 AM To: Asim Roy ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Asim, I sit here with my coffee neuron and my cup neuron creating a cup of coffee, and my table neuron holding the cup and coffee neurons representing a cup of coffee. On the table, on the floor, noting of course the cup of coffee is not on the floor but on the table. So this is a network of coffee, cups, tables, maybe coded by another neuron in a hierarchical network called morning coffee? Well this is the typical problem with localist model of the brain. How would it work exactly? So far just representing a single episodic event requires hierarchical network of neurons that generalize to what exactly? Do we need many such networks to represent all episodic memories of having a cup of coffee? This is similar to the problem with the socalled "fusiform face area" which with some proper controls doesn't really exist (cf Hanson 2022). What would a small blueberry are in fusiform gyrus be doing with all those faces we recognize? A brain face cloud server? Other accounts show that FG, LO, IN and PFC are at least involved in a network of areas (not neurons) in recognizing a face. But computationally what's the actual underlying functions? The shift to "neurons represent symbols' is not helpful to explain meaning. Remember, we have billions of neurons and 100 trillions connections between them. I'm not saying symbols aren't important and Harnad's Grounding arguments are becoming more critical (and need to be made more explicit) as LLMs start to be embedded in robots. Therefore an arguably simpler task, is to identify "symbols" in LLMs since we have no idea how they work either. Are symbols in their "neurons"? Cheers Stephen On 8/24/26 22:56, Asim Roy wrote: 1. Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? 2. Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? 3. Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195?223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -- [cid:image003.jpg at 01DD3625.F5884020] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image003.jpg Type: image/jpeg Size: 57383 bytes Desc: image003.jpg URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image004.jpg Type: image/jpeg Size: 45946 bytes Desc: image004.jpg URL: From lorincz at inf.elte.hu Sat Aug 29 02:00:24 2026 From: lorincz at inf.elte.hu (=?utf-8?B?TMWRcmluY3ogQW5kcsOhcw==?=) Date: Sat, 29 Aug 2026 06:00:24 +0000 Subject: Connectionists: AI@50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> Message-ID: Dear Steve: Beyond the problem of category proliferation ?? if the input has \(d\) features and ART has learned \(K\) categories, a straightforward category search requires approximately \(O(Kd)\) computation and \(O(Kd)\) memory. This is not intrinsically exponential in \(d\), but \(K\) may approach the number of individual experiences. Worse, if a task requires distinguishing combinations of \(m\) approximately independent factors, each having \(r\) possible values, the number of required conjunctive categories may, in the worst case, grow as \(r^m\) ?? I am particularly concerned about the concept of resonance. Precise computation does not necessarily require global synchrony, while constructing a coherent representation of a dynamic world requires both predictive and postdictive processing to compensate for heterogeneous delays. If ?resonance? merely means transient, local, and delay-tolerant recurrent agreement, I see no fundamental incompatibility. However, if it means a globally synchronized or settled state that produces perception or consciousness, then I have a serious timing objection. Such a state appears incompatible with a distributed, unsynchronized system that nevertheless represents the world with millisecond precision by means of delay compensation, prediction, and postdiction; global resonance would simply take too long, I am afraid. Best, Andras ------------------------------------ Andras Lorincz Fellow of the European Association for Artificial Intelligence ELLIS Member Department of Artificial Intelligence Faculty of Informatics Eotvos Lorand University Budapest, Hungary ________________________________ From: Connectionists on behalf of Grossberg, Stephen Sent: Friday, August 28, 2026 17:00 To: Asim Roy ; Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Asim, Thanks for bringing up the DARPA Explainable AI program. I was invited to submit an article to a Special Issue about Explainable AI that was published when the DARPA program was announced. The editors of the Special Issue knew that Adaptive Resonance Theory, or ART, models automatically discover and incrementally learn over multiple learning trials to pay attention to the predictive, or explainable, critical feature patterns that control successful predictions and are used in learning by the adaptive weights in bottom-up adaptive filters and top-down expectations. ART does this AUTOMATICALLY. You seem to have to use an external teacher to do it. Here is the article: Grossberg, S. (2020). A path towards Explainable AI and autonomous adaptive intelligence: Deep Learning, Adaptive Resonance, and models of perception, emotion, and action. Frontiers in Neurobotics, June 25, 2020. https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2020.00036/full The article?s Abstract illustrates its explanatory range [boldface mine]: * "Biological neural network models whereby brains make minds help to understand autonomous adaptive intelligence. This article summarizes why the dynamics and emergent properties of such models for perception, cognition, emotion, and action are explainable, and thus amenable to being confidently implemented in large-scale applications. Key to their explainability is how these models combine fast activations, or short-term memory (STM) traces, and learned weights, or long-term memory (LTM) traces. Visual and auditory perceptual models have explainable conscious STM representations of visual surfaces and auditory streams in surface-shroud resonances and stream-shroud resonances, respectively. Deep Learning is often used to classify data. However, Deep Learning can experience catastrophic forgetting: At any stage of learning, an unpredictable part of its memory can collapse. Even if it makes some accurate classifications, they are not explainable and thus cannot be used with confidence. Deep Learning shares these problems with the back propagation algorithm, whose computational problems due to non-local weight transport during mismatch learning were described in the 1980s. Deep Learning became popular after very fast computers and huge online databases became available that enabled new applications despite these problems. Adaptive Resonance Theory, or ART, algorithms overcome the computational problems of back propagation and Deep Learning. ART is a self-organizing production system that incrementally learns, using arbitrary combinations of unsupervised and supervised learning and only locally computable quantities, to rapidly classify large non-stationary databases without experiencing catastrophic forgetting. ART classifications and predictions are explainable using the attended critical feature patterns in STM on which they build. The LTM adaptive weights of the fuzzy ARTMAP algorithm induce fuzzy IF-THEN rules that explain what feature combinations predict successful outcomes. ART has been successfully used in multiple large-scale real-world applications, including remote sensing, medical database prediction, and social media data clustering. Also explainable are the MOTIVATOR model of reinforcement learning and cognitive-emotional interactions, and the VITE, DIRECT, DIVA, and SOVEREIGN models for reaching, speech production, spatial navigation, and autonomous adaptive intelligence. These biological models exemplify complementary computing, and use local laws for match learning and mismatch learning that avoid the problems of Deep Learning." Chapter 5 in my 2021 Magnum Opus CONSCIOUS MIND, RESONANT BRAIN: HOW EACH BRAIN MAKES A MIND https://www.amazon.com/dp/0190070552?lv=shuf&channelId=500&plpRedirect=mhFallback provides a self-contained and non-technical overview and synthesis of my results about ART and Explainable AI. The other chapters summarize neural network models of the main processes whereby our brains make our conscious minds in healthy individuals and clinical patients. Best, Steve Stephen Grossberg Wang Professor of Cognitive and Neural Systems Director, Center for Adaptive Systems Emeritus Professor of Mathematics & Statistics, Psychological & Brain Sciences, and Biomedical Engineering Boston University sites.bu.edu/steveg/ steve at bu.edu http://en.wikipedia.org/wiki/Stephen_Grossberg http://scholar.google.com/citations?user=3BIV70wAAAAJ&hl=en https://sites.bu.edu/steveg/files/2021/08/Grossberg-CV-8-14-21.pdf https://youtu.be/9n5AnvFur7I https://www.youtube.com/watch?v=_hBye6JQCh4 https://www.amazon.com/Conscious-Mind-Resonant-Brain-Makes/dp/0190070552 https://www.amazon.com/Your-Creative-Brain-Consciously-Experience/dp/0198965370 From: Connectionists on behalf of Asim Roy Date: Friday, August 28, 2026 at 3:51?AM To: Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Stephen, You raise important issues. There are many ways of dealing with them. Here?s one of the ways. The image below shows what DARPA wanted for Explainable AI. Essentially, find the parts of an object for its prediction. So, for a cat, verify some of its unique parts like the fur, whiskers, and claws. Thus, in our explainable models, we ?teach? an object detection model about these parts. What we get as outputs, for a cat, are predictions for the cat and its parts. There?s a layer of logic that uses these outputs and verifies the existence of the parts before ?finally? predicting that there?s a cat in the image. In essence, it?s neuro-symbolic. One can do the symbolic part with a graphical model too. We can also predict that there?s a cat even if we don?t ?see? all of the parts, such as when we just see its face and not the rest of the body. And, in a similar way, your ?morning coffee? concept can be built from its parts that you mention. There?s a long tradition in computer vision of finding parts of objects inside the model. Our initial conception was that the higher-level filters in a CNN correspond to certain abstractions such as a nose, eyes, ears for a human. That was the distributed representation but based on lower-level abstractions (parts of objects). But years of research showed that one could not find these kinds of lower-level abstractions in standard CNN models. So, what do you do? Well, one way is to force some of the filters to correspond to these parts. There?s a long history to this body of work in computer vision, including that of Hinton, to create abstractions within a model. Both ways, we are ?teaching? the models about ?low-level abstraction.? In a way, in both ways, it?s a hierarchical system. In both ways, we create a neuro-symbolic system. By the way, abstractions are generalizations, they don?t code specific episodic memories. Your ?table,? ?coffee,? ?cup? and all that are abstract notions. Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [A screenshot of a computer AI-generated content may be incorrect.] From: Stephen Jos? Hanson Sent: Thursday, August 27, 2026 4:26 AM To: Asim Roy ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Asim, I sit here with my coffee neuron and my cup neuron creating a cup of coffee, and my table neuron holding the cup and coffee neurons representing a cup of coffee. On the table, on the floor, noting of course the cup of coffee is not on the floor but on the table. So this is a network of coffee, cups, tables, maybe coded by another neuron in a hierarchical network called morning coffee? Well this is the typical problem with localist model of the brain. How would it work exactly? So far just representing a single episodic event requires hierarchical network of neurons that generalize to what exactly? Do we need many such networks to represent all episodic memories of having a cup of coffee? This is similar to the problem with the socalled "fusiform face area" which with some proper controls doesn't really exist (cf Hanson 2022). What would a small blueberry are in fusiform gyrus be doing with all those faces we recognize? A brain face cloud server? Other accounts show that FG, LO, IN and PFC are at least involved in a network of areas (not neurons) in recognizing a face. But computationally what's the actual underlying functions? The shift to "neurons represent symbols' is not helpful to explain meaning. Remember, we have billions of neurons and 100 trillions connections between them. I'm not saying symbols aren't important and Harnad's Grounding arguments are becoming more critical (and need to be made more explicit) as LLMs start to be embedded in robots. Therefore an arguably simpler task, is to identify "symbols" in LLMs since we have no idea how they work either. Are symbols in their "neurons"? Cheers Stephen On 8/24/26 22:56, Asim Roy wrote: 1. Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? 2. Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? 3. Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195?223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -- [cid:image003.jpg at 01DD3625.F5884020] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image003.jpg Type: image/jpeg Size: 57383 bytes Desc: image003.jpg URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image004.jpg Type: image/jpeg Size: 45946 bytes Desc: image004.jpg URL: From pinghsieh at nycu.edu.tw Sat Aug 29 05:55:47 2026 From: pinghsieh at nycu.edu.tw (Ping-Chun Hsieh) Date: Sat, 29 Aug 2026 17:55:47 +0800 Subject: Connectionists: [CFP] NeurIPS 2026 Workshop: From Pretrained Representations to Acting Agents (Deadline: Sept 5, 2026 AoE) Message-ID: Dear All, We are excited to announce the PTA: From Pretrained Representations to Acting Agents workshop at NeurIPS 2026 in Sydney, Australia! ? We?re calling for submissions! ? Paper submission deadline: Sept 5, 2026 (AoE) ? Website: https://ptaworkshop.github.io/ ? Submission link: https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PTA Our goal is to explore how to obtain actionable pretrained representations and align them with control agents at test time?bridging pretraining and test-time decision making. Call for Papers: - Specific topics include, but are not limited to: 1. What makes a pretrained representation actionable? 2. How to align with and use pretrained representations at test time? 3. How to adapt pretrained knowledge to new tasks? 4. How to evaluate actionable representations? 5. What can go wrong when pretrained agents act? - We welcome both short papers (4 pages) and long papers (up to 9 pages), formatted using the NeurIPS template and submitted through OpenReview. - All accepted contributions will be presented during the poster sessions. - A select number of submissions will also be invited for contributed talks. - Submissions are non-archival and may be under review or concurrently submitted elsewhere. We especially encourage ongoing and unpublished work. If you are excited about connecting representation learning, RL, planning, and test-time decision making, we hope you?ll join us in Sydney! ? Best regards, The Organizing Team Ping-Chun Hsieh (NYCU), Kuang-Huei Lee (Google DeepMind), Bo Dai (Georgia Tech), Yen-Ling Kuo (UVA), Georgia Chalvatzaki (TU Darmstadt), Karen Leung (UW / NVIDIA), Co Yong (NTU), Claas Voelcker (UT Austin) -- Ping-Chun Hsieh (???) Associate Professor Department of Computer Science National Yang Ming Chiao Tung University https://pinghsieh.github.io/ -------------- next part -------------- An HTML attachment was scrubbed... URL: From ASIM.ROY at asu.edu Sat Aug 29 05:08:17 2026 From: ASIM.ROY at asu.edu (Asim Roy) Date: Sat, 29 Aug 2026 09:08:17 +0000 Subject: Connectionists: AI@50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> Message-ID: Dear Steve, What DARPA seemed to want at that time for computer vision/object recognition is object prediction based on part verification ? if a cat, have you verified its unique parts? That?s the kind of explanation they wanted. ?Teaching? an object recognition system about these parts (to generalize, parts are just subconcepts) is not inconsistent with the way we teach humans. For example, a one- or two-year-old recognizes a car. But we teach them about parts much later - such as what a tire is, a door, steering wheel and so on. How a person?s internal biological model is adjusted from these ?part teachings? I am not sure, but the identification and teaching (labeling) about parts is very much a part of human learning. One might say that LLMs automate this type of learning, but we do provide labels in the descriptions of objects. So, that?s ?teaching.? When we started with multilayered models, the assumption was that at higher levels, there would be part abstractions such as those shown in the figure below - the dog face, legs, tail and so on. The problem was, we couldn?t find these abstractions at higher levels of the network. Even if you found them, we would need to assign labels or names to them ? a dog face, legs, tail and so on. Once you get to assigning these labels, that?s ?teaching.? There is a body of work in computer vision trying to build these abstractions inside CNN models. In our method, we also ?teach? our deep learning models, but in a different way. In general, ?teaching? would be necessary whichever way you do it. We call our approach ?Explanation First, Model Next.? We have reversed the process of model building. Our testing shows gain in overall accuracy compared to some standard object detection models. This DARPA style approach also enables resistance against adversarial attacks. So substantial benefits with the Explanation First approach. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [cid:image003.png at 01DD3758.6623B9D0] From: Grossberg, Stephen Sent: Friday, August 28, 2026 8:01 AM To: Asim Roy ; Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen Subject: Re: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Asim, Thanks for bringing up the DARPA Explainable AI program. I was invited to submit an article to a Special Issue about Explainable AI that was published when the DARPA program was announced. The editors of the Special Issue knew that Adaptive Resonance Theory, or ART, models automatically discover and incrementally learn over multiple learning trials to pay attention to the predictive, or explainable, critical feature patterns that control successful predictions and are used in learning by the adaptive weights in bottom-up adaptive filters and top-down expectations. ART does this AUTOMATICALLY. You seem to have to use an external teacher to do it. Here is the article: Grossberg, S. (2020). A path towards Explainable AI and autonomous adaptive intelligence: Deep Learning, Adaptive Resonance, and models of perception, emotion, and action. Frontiers in Neurobotics, June 25, 2020. https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2020.00036/full The article?s Abstract illustrates its explanatory range [boldface mine]: * "Biological neural network models whereby brains make minds help to understand autonomous adaptive intelligence. This article summarizes why the dynamics and emergent properties of such models for perception, cognition, emotion, and action are explainable, and thus amenable to being confidently implemented in large-scale applications. Key to their explainability is how these models combine fast activations, or short-term memory (STM) traces, and learned weights, or long-term memory (LTM) traces. Visual and auditory perceptual models have explainable conscious STM representations of visual surfaces and auditory streams in surface-shroud resonances and stream-shroud resonances, respectively. Deep Learning is often used to classify data. However, Deep Learning can experience catastrophic forgetting: At any stage of learning, an unpredictable part of its memory can collapse. Even if it makes some accurate classifications, they are not explainable and thus cannot be used with confidence. Deep Learning shares these problems with the back propagation algorithm, whose computational problems due to non-local weight transport during mismatch learning were described in the 1980s. Deep Learning became popular after very fast computers and huge online databases became available that enabled new applications despite these problems. Adaptive Resonance Theory, or ART, algorithms overcome the computational problems of back propagation and Deep Learning. ART is a self-organizing production system that incrementally learns, using arbitrary combinations of unsupervised and supervised learning and only locally computable quantities, to rapidly classify large non-stationary databases without experiencing catastrophic forgetting. ART classifications and predictions are explainable using the attended critical feature patterns in STM on which they build. The LTM adaptive weights of the fuzzy ARTMAP algorithm induce fuzzy IF-THEN rules that explain what feature combinations predict successful outcomes. ART has been successfully used in multiple large-scale real-world applications, including remote sensing, medical database prediction, and social media data clustering. Also explainable are the MOTIVATOR model of reinforcement learning and cognitive-emotional interactions, and the VITE, DIRECT, DIVA, and SOVEREIGN models for reaching, speech production, spatial navigation, and autonomous adaptive intelligence. These biological models exemplify complementary computing, and use local laws for match learning and mismatch learning that avoid the problems of Deep Learning." Chapter 5 in my 2021 Magnum Opus CONSCIOUS MIND, RESONANT BRAIN: HOW EACH BRAIN MAKES A MIND https://www.amazon.com/dp/0190070552?lv=shuf&channelId=500&plpRedirect=mhFallback provides a self-contained and non-technical overview and synthesis of my results about ART and Explainable AI. The other chapters summarize neural network models of the main processes whereby our brains make our conscious minds in healthy individuals and clinical patients. Best, Steve Stephen Grossberg Wang Professor of Cognitive and Neural Systems Director, Center for Adaptive Systems Emeritus Professor of Mathematics & Statistics, Psychological & Brain Sciences, and Biomedical Engineering Boston University sites.bu.edu/steveg/ steve at bu.edu http://en.wikipedia.org/wiki/Stephen_Grossberg http://scholar.google.com/citations?user=3BIV70wAAAAJ&hl=en https://sites.bu.edu/steveg/files/2021/08/Grossberg-CV-8-14-21.pdf https://youtu.be/9n5AnvFur7I https://www.youtube.com/watch?v=_hBye6JQCh4 https://www.amazon.com/Conscious-Mind-Resonant-Brain-Makes/dp/0190070552 https://www.amazon.com/Your-Creative-Brain-Consciously-Experience/dp/0198965370 From: Connectionists > on behalf of Asim Roy > Date: Friday, August 28, 2026 at 3:51?AM To: Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: AI at 50 videos... Stephen, You raise important issues. There are many ways of dealing with them. Here?s one of the ways. The image below shows what DARPA wanted for Explainable AI. Essentially, find the parts of an object for its prediction. So, for a cat, verify some of its unique parts like the fur, whiskers, and claws. Thus, in our explainable models, we ?teach? an object detection model about these parts. What we get as outputs, for a cat, are predictions for the cat and its parts. There?s a layer of logic that uses these outputs and verifies the existence of the parts before ?finally? predicting that there?s a cat in the image. In essence, it?s neuro-symbolic. One can do the symbolic part with a graphical model too. We can also predict that there?s a cat even if we don?t ?see? all of the parts, such as when we just see its face and not the rest of the body. And, in a similar way, your ?morning coffee? concept can be built from its parts that you mention. There?s a long tradition in computer vision of finding parts of objects inside the model. Our initial conception was that the higher-level filters in a CNN correspond to certain abstractions such as a nose, eyes, ears for a human. That was the distributed representation but based on lower-level abstractions (parts of objects). But years of research showed that one could not find these kinds of lower-level abstractions in standard CNN models. So, what do you do? Well, one way is to force some of the filters to correspond to these parts. There?s a long history to this body of work in computer vision, including that of Hinton, to create abstractions within a model. Both ways, we are ?teaching? the models about ?low-level abstraction.? In a way, in both ways, it?s a hierarchical system. In both ways, we create a neuro-symbolic system. By the way, abstractions are generalizations, they don?t code specific episodic memories. Your ?table,? ?coffee,? ?cup? and all that are abstract notions. Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [A screenshot of a computer AI-generated content may be incorrect.] From: Stephen Jos? Hanson > Sent: Thursday, August 27, 2026 4:26 AM To: Asim Roy >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Asim, I sit here with my coffee neuron and my cup neuron creating a cup of coffee, and my table neuron holding the cup and coffee neurons representing a cup of coffee. On the table, on the floor, noting of course the cup of coffee is not on the floor but on the table. So this is a network of coffee, cups, tables, maybe coded by another neuron in a hierarchical network called morning coffee? Well this is the typical problem with localist model of the brain. How would it work exactly? So far just representing a single episodic event requires hierarchical network of neurons that generalize to what exactly? Do we need many such networks to represent all episodic memories of having a cup of coffee? This is similar to the problem with the socalled "fusiform face area" which with some proper controls doesn't really exist (cf Hanson 2022). What would a small blueberry are in fusiform gyrus be doing with all those faces we recognize? A brain face cloud server? Other accounts show that FG, LO, IN and PFC are at least involved in a network of areas (not neurons) in recognizing a face. But computationally what's the actual underlying functions? The shift to "neurons represent symbols' is not helpful to explain meaning. Remember, we have billions of neurons and 100 trillions connections between them. I'm not saying symbols aren't important and Harnad's Grounding arguments are becoming more critical (and need to be made more explicit) as LLMs start to be embedded in robots. Therefore an arguably simpler task, is to identify "symbols" in LLMs since we have no idea how they work either. Are symbols in their "neurons"? Cheers Stephen On 8/24/26 22:56, Asim Roy wrote: 1. Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? 2. Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? 3. Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195?223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -- [cid:image002.jpg at 01DD3755.3D4B3630] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image001.jpg Type: image/jpeg Size: 45946 bytes Desc: image001.jpg URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image002.jpg Type: image/jpeg Size: 57383 bytes Desc: image002.jpg URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image003.png Type: image/png Size: 333780 bytes Desc: image003.png URL: From steve at bu.edu Sat Aug 29 09:24:43 2026 From: steve at bu.edu (Grossberg, Stephen) Date: Sat, 29 Aug 2026 13:24:43 +0000 Subject: Connectionists: AI@50 videos...Self-organized learning of objects and their parts In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> Message-ID: Dear Asim, Models of the ARTMAP family (fuzzy ARTMAP, distributed ARTMAP, Gaussian ARTMAP, etc.) are perhaps the simplest kind of self-organizing neural model that can use a teacher (typically the environment, but it can also be a human) to learn to recognize objects, their parts, and their names. The work that my colleagues and I have published has gone far beyond ARTMAP. We have, in a series of articles that can be downloaded from sites.bu.edu/steveg, shown how humans can learn view-, position-, and size-invariant object representations, and use them to carry out actions that achieve valued goals. Our models show how both view-specific and view-invariant representations of an object (say, a particular view of your mother?s face AND an invariant one; namely, that she has a face) can simultaneously guide recognition of mom AND specific actions in response to where the view of her face is looking. The ?views? are ?parts?. I will list some of the relevant article titles and their urls here, for your convenience. Later articles tend to model ever more challenging aspects of self-organizing object recognition and action. DARPA gave us large grants over the years to do this work. Carpenter, G.A., Grossberg, S., and Mehanian, C. (1989). Invariant recognition of cluttered scenes by a self- organizing ART architecture: CORT-X boundary segmentation. Neural Networks, 2, 169-181. https://sites.bu.edu/steveg/files/2016/06/CarGroMeh1989NN.pdf Grossberg, S. and Wyse, L. (1991). Invariant recognition of cluttered scenes by a self-organizing ART architecture: Figure-ground separation. Neural Networks, 1991, 4, 723-742. https://sites.bu.edu/steveg/files/2016/06/GroWyse1991NN.pdf Fazl, A., Grossberg, S., and Mingolla, E. (2009). View-invariant object category learning, recognition, and search: How spatial and object attention are coordinated using surface-based attentional shrouds. Cognitive Psychology, 58, 1-48. https://sites.bu.edu/steveg/files/2016/06/FazGroMin2008.pdf Grossberg, S., and Vladusich, T. (2010). How do children learn to follow gaze, share joint attention, imitate their teachers, and use tools during social interactions? Neural Networks, 23, 940-965. https://sites.bu.edu/steveg/files/2016/06/GrossbergVladusichNN2010.pdf Cao, Y., Grossberg, S., and Markowitz, J. (2011). How does the brain rapidly learn and reorganize view- and positionally-invariant object representations in inferior temporal cortex? Neural Networks, 24, 1050-1061. https://sites.bu.edu/steveg/files/2016/06/NN2853.pdf Grossberg, S., Markowitz, J., and Cao, Y. (2011). On the road to invariant recognition: Explaining tradeoff and morph properties of cells in inferotemporal cortex using multiple-scale task-sensitive attentive learning. Neural Networks, 24, 1036-1049. https://sites.bu.edu/steveg/files/2016/06/GroMarCao2011TR.pdf Grossberg, S., Srinivasan, K., and Yazdabakhsh, A. (2011). On the road to invariant object recognition: How cortical area V2 transforms absolute to relative disparity during 3D vision. Neural Networks, 24, 686-692. https://sites.bu.edu/steveg/files/2016/06/GroSriYaz2011TR.pdf Chang, H.-C., Grossberg, S., and Cao, Y. (2014) Where?s Waldo? How perceptual cognitive, and emotional brain processes cooperate during learning to categorize and find desired objects in a cluttered scene. Frontiers in Integrative Neuroscience, doi: 10.3389/fnint.2014.0043, https://www.frontiersin.org/journals/integrative-neuroscience/articles/10.3389/fnint.2014.00043/full Grossberg, S., Srinivasan, K., and Yazdanbakhsh, A. (2014). Binocular fusion and invariant category learning due to predictive remapping during scanning of a depthful scene with eye movements. Frontiers in Psychology: Perception Science, doi: 10.3389/fpsyg.2014.01457. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.01457/full Grossberg, S. (2019). The embodied brain of SOVEREIGN2: From space-variant conscious percepts during visual search and navigation to learning invariant object categories and cognitive-emotional plans for acquiring valued goals. Frontiers in Computational Neuroscience. Published online: June 25, 2019. https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2019.00036/full Best, Steve Get Outlook for Mac From: Asim Roy Date: Saturday, August 29, 2026 at 5:09?AM To: Grossberg, Stephen ; Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Steve, What DARPA seemed to want at that time for computer vision/object recognition is object prediction based on part verification ? if a cat, have you verified its unique parts? That?s the kind of explanation they wanted. ?Teaching? an object recognition system about these parts (to generalize, parts are just subconcepts) is not inconsistent with the way we teach humans. For example, a one- or two-year-old recognizes a car. But we teach them about parts much later - such as what a tire is, a door, steering wheel and so on. How a person?s internal biological model is adjusted from these ?part teachings? I am not sure, but the identification and teaching (labeling) about parts is very much a part of human learning. One might say that LLMs automate this type of learning, but we do provide labels in the descriptions of objects. So, that?s ?teaching.? When we started with multilayered models, the assumption was that at higher levels, there would be part abstractions such as those shown in the figure below - the dog face, legs, tail and so on. The problem was, we couldn?t find these abstractions at higher levels of the network. Even if you found them, we would need to assign labels or names to them ? a dog face, legs, tail and so on. Once you get to assigning these labels, that?s ?teaching.? There is a body of work in computer vision trying to build these abstractions inside CNN models. In our method, we also ?teach? our deep learning models, but in a different way. In general, ?teaching? would be necessary whichever way you do it. We call our approach ?Explanation First, Model Next.? We have reversed the process of model building. Our testing shows gain in overall accuracy compared to some standard object detection models. This DARPA style approach also enables resistance against adversarial attacks. So substantial benefits with the Explanation First approach. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [cid:image003.png at 01DD3758.6623B9D0] From: Grossberg, Stephen Sent: Friday, August 28, 2026 8:01 AM To: Asim Roy ; Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen Subject: Re: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Asim, Thanks for bringing up the DARPA Explainable AI program. I was invited to submit an article to a Special Issue about Explainable AI that was published when the DARPA program was announced. The editors of the Special Issue knew that Adaptive Resonance Theory, or ART, models automatically discover and incrementally learn over multiple learning trials to pay attention to the predictive, or explainable, critical feature patterns that control successful predictions and are used in learning by the adaptive weights in bottom-up adaptive filters and top-down expectations. ART does this AUTOMATICALLY. You seem to have to use an external teacher to do it. Here is the article: Grossberg, S. (2020). A path towards Explainable AI and autonomous adaptive intelligence: Deep Learning, Adaptive Resonance, and models of perception, emotion, and action. Frontiers in Neurobotics, June 25, 2020. https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2020.00036/full The article?s Abstract illustrates its explanatory range [boldface mine]: * "Biological neural network models whereby brains make minds help to understand autonomous adaptive intelligence. This article summarizes why the dynamics and emergent properties of such models for perception, cognition, emotion, and action are explainable, and thus amenable to being confidently implemented in large-scale applications. Key to their explainability is how these models combine fast activations, or short-term memory (STM) traces, and learned weights, or long-term memory (LTM) traces. Visual and auditory perceptual models have explainable conscious STM representations of visual surfaces and auditory streams in surface-shroud resonances and stream-shroud resonances, respectively. Deep Learning is often used to classify data. However, Deep Learning can experience catastrophic forgetting: At any stage of learning, an unpredictable part of its memory can collapse. Even if it makes some accurate classifications, they are not explainable and thus cannot be used with confidence. Deep Learning shares these problems with the back propagation algorithm, whose computational problems due to non-local weight transport during mismatch learning were described in the 1980s. Deep Learning became popular after very fast computers and huge online databases became available that enabled new applications despite these problems. Adaptive Resonance Theory, or ART, algorithms overcome the computational problems of back propagation and Deep Learning. ART is a self-organizing production system that incrementally learns, using arbitrary combinations of unsupervised and supervised learning and only locally computable quantities, to rapidly classify large non-stationary databases without experiencing catastrophic forgetting. ART classifications and predictions are explainable using the attended critical feature patterns in STM on which they build. The LTM adaptive weights of the fuzzy ARTMAP algorithm induce fuzzy IF-THEN rules that explain what feature combinations predict successful outcomes. ART has been successfully used in multiple large-scale real-world applications, including remote sensing, medical database prediction, and social media data clustering. Also explainable are the MOTIVATOR model of reinforcement learning and cognitive-emotional interactions, and the VITE, DIRECT, DIVA, and SOVEREIGN models for reaching, speech production, spatial navigation, and autonomous adaptive intelligence. These biological models exemplify complementary computing, and use local laws for match learning and mismatch learning that avoid the problems of Deep Learning." Chapter 5 in my 2021 Magnum Opus CONSCIOUS MIND, RESONANT BRAIN: HOW EACH BRAIN MAKES A MIND https://www.amazon.com/dp/0190070552?lv=shuf&channelId=500&plpRedirect=mhFallback provides a self-contained and non-technical overview and synthesis of my results about ART and Explainable AI. The other chapters summarize neural network models of the main processes whereby our brains make our conscious minds in healthy individuals and clinical patients. Best, Steve Stephen Grossberg Wang Professor of Cognitive and Neural Systems Director, Center for Adaptive Systems Emeritus Professor of Mathematics & Statistics, Psychological & Brain Sciences, and Biomedical Engineering Boston University sites.bu.edu/steveg/ steve at bu.edu http://en.wikipedia.org/wiki/Stephen_Grossberg http://scholar.google.com/citations?user=3BIV70wAAAAJ&hl=en https://sites.bu.edu/steveg/files/2021/08/Grossberg-CV-8-14-21.pdf https://youtu.be/9n5AnvFur7I https://www.youtube.com/watch?v=_hBye6JQCh4 https://www.amazon.com/Conscious-Mind-Resonant-Brain-Makes/dp/0190070552 https://www.amazon.com/Your-Creative-Brain-Consciously-Experience/dp/0198965370 From: Connectionists > on behalf of Asim Roy > Date: Friday, August 28, 2026 at 3:51?AM To: Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: AI at 50 videos... Stephen, You raise important issues. There are many ways of dealing with them. Here?s one of the ways. The image below shows what DARPA wanted for Explainable AI. Essentially, find the parts of an object for its prediction. So, for a cat, verify some of its unique parts like the fur, whiskers, and claws. Thus, in our explainable models, we ?teach? an object detection model about these parts. What we get as outputs, for a cat, are predictions for the cat and its parts. There?s a layer of logic that uses these outputs and verifies the existence of the parts before ?finally? predicting that there?s a cat in the image. In essence, it?s neuro-symbolic. One can do the symbolic part with a graphical model too. We can also predict that there?s a cat even if we don?t ?see? all of the parts, such as when we just see its face and not the rest of the body. And, in a similar way, your ?morning coffee? concept can be built from its parts that you mention. There?s a long tradition in computer vision of finding parts of objects inside the model. Our initial conception was that the higher-level filters in a CNN correspond to certain abstractions such as a nose, eyes, ears for a human. That was the distributed representation but based on lower-level abstractions (parts of objects). But years of research showed that one could not find these kinds of lower-level abstractions in standard CNN models. So, what do you do? Well, one way is to force some of the filters to correspond to these parts. There?s a long history to this body of work in computer vision, including that of Hinton, to create abstractions within a model. Both ways, we are ?teaching? the models about ?low-level abstraction.? In a way, in both ways, it?s a hierarchical system. In both ways, we create a neuro-symbolic system. By the way, abstractions are generalizations, they don?t code specific episodic memories. Your ?table,? ?coffee,? ?cup? and all that are abstract notions. Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [A screenshot of a computer AI-generated content may be incorrect.] From: Stephen Jos? Hanson > Sent: Thursday, August 27, 2026 4:26 AM To: Asim Roy >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Asim, I sit here with my coffee neuron and my cup neuron creating a cup of coffee, and my table neuron holding the cup and coffee neurons representing a cup of coffee. On the table, on the floor, noting of course the cup of coffee is not on the floor but on the table. So this is a network of coffee, cups, tables, maybe coded by another neuron in a hierarchical network called morning coffee? Well this is the typical problem with localist model of the brain. How would it work exactly? So far just representing a single episodic event requires hierarchical network of neurons that generalize to what exactly? Do we need many such networks to represent all episodic memories of having a cup of coffee? This is similar to the problem with the socalled "fusiform face area" which with some proper controls doesn't really exist (cf Hanson 2022). What would a small blueberry are in fusiform gyrus be doing with all those faces we recognize? A brain face cloud server? Other accounts show that FG, LO, IN and PFC are at least involved in a network of areas (not neurons) in recognizing a face. But computationally what's the actual underlying functions? The shift to "neurons represent symbols' is not helpful to explain meaning. Remember, we have billions of neurons and 100 trillions connections between them. I'm not saying symbols aren't important and Harnad's Grounding arguments are becoming more critical (and need to be made more explicit) as LLMs start to be embedded in robots. Therefore an arguably simpler task, is to identify "symbols" in LLMs since we have no idea how they work either. Are symbols in their "neurons"? Cheers Stephen On 8/24/26 22:56, Asim Roy wrote: 1. Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? 2. Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? 3. Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195?223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -- [cid:image002.jpg at 01DD3755.3D4B3630] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image001.jpg Type: image/jpeg Size: 45946 bytes Desc: image001.jpg URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image002.jpg Type: image/jpeg Size: 57383 bytes Desc: image002.jpg URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image003.png Type: image/png Size: 333780 bytes Desc: image003.png URL: From david at irdta.eu Sat Aug 29 10:03:08 2026 From: david at irdta.eu (David Silva - IRDTA) Date: Sat, 29 Aug 2026 16:03:08 +0200 (CEST) Subject: Connectionists: GeMAIHc 2027: early registration September 18 Message-ID: <630222850.18007.1788012188728@webmail.strato.com> ******************************************************** 1st INTERNATIONAL SCHOOL ON GENERATIVE & MULTIMODAL AI FOR HEALTHCARE GeMAIHc 2027 Milan, Italy February 1-5, 2027 https://gemaihc.irdta.eu/2027/ ******************************************************** Co-organized by: Human Technopole IRDTA ? Institute for Research Development, Training and Advice ****************************************************** Early registration: September 18, 2026 ****************************************************** SCOPE AI is rapidly redefining biomedical research, clinical decision-making, and healthcare delivery worldwide. GeMAIHc 2027 will provide participants with a comprehensive overview of state-of-the-art AI methodologies. Topics will include generative modelling, multimodal data integration and augmentation, large language models and clinical NLP, medical image synthesis and analysis, as well as evaluation, robustness, validation, ethical considerations, and regulatory aspects of AI systems in healthcare. GeMAIHc 2027 aims to become a leading international forum at the intersection of generative AI, multimodal learning, and healthcare innovation. The program will emphasize both methodological foundations and real-world clinical applications, combining theoretical insights with practical perspectives. This interdisciplinary event will feature 18 monographic three-hour courses, 2 keynote lectures, 3 scientific sessions, 1 symposium, and a hackathon. Leading academics and industry pioneers will share their expertise and perspectives with attendees. In-person interaction and networking will be central components of the event, while full remote participation will also be possible. ADDRESSED TO The program is open to PhD students and postdoctoral researchers in AI, data science, and biomedical disciplines, clinicians and healthcare professionals, industry practitioners and innovators, and policymakers and stakeholders in digital health. There are no formal academic prerequisites for participation. Researchers and professionals at all career stages are welcome. An international audience is expected, with backgrounds spanning computer science, medicine, engineering, mathematics, statistics, physics, and the social sciences. GeMAIHc 2027 will offer a unique opportunity to gain in-depth knowledge of a rapidly evolving field, to interact with experts across disciplines, to establish collaborations and expand professional networks, and to explore how AI is reshaping the future of healthcare. VENUE GeMAIHc 2027 will take place in Milan, one of Europe?s leading international centers for science, industry, fashion, and finance. The venue will be: Human Technopole Viale Rita Levi-Montalcini, 1 20157 Milan, Italy https://humantechnopole.it/en/ STRUCTURE Three parallel courses will run throughout the event. Participants will be free to choose the sessions they wish to attend and may move between courses at any time. A symposium will offer participants and companies the opportunity to present ongoing research or industrial developments through 10-minute oral presentations. The school will include a hackathon, during which participants will work in teams to tackle challenges in generative and multimodal AI for healthcare. All lectures will be video recorded and made available to participants for 45 days after the event. Full live online participation will be possible. Nevertheless, the organizers emphasize the importance of in-person interaction and networking in research training events of this kind. KEYNOTE SPEAKERS Deborah Estrin (Cornell University), Transforming Longitudinal Care with Digital Biomarkers and Therapeutics Lyle Palmer (Adelaide University), Deep Learning in Medicine: Some Lessons from Epidemiology and Genomics PROFESSORS AND COURSES Pierre Baldi (University of California Irvine), [introductory/advanced] The AI-driven Healthcare of the Future David Buckeridge (McGill University), [intermediate] Multi-Modal Data Analysis in Population and Public Health Alejandro Frangi (University of Manchester), [intermediate/advanced] Virtual Patient Populations for In Silico Trials Using Generative AI Based on Multimodal Real-World Data Charles Friedman (University of Michigan), [introductory] Synergizing AI and Learning Health Systems to Transform Health Judy Wawira Gichoya (Emory University), [introductory/advanced] Combination in a Hands on Lab for Harmonizing Radiology Datasets Maryellen Giger (University of Chicago), [introductory/intermediate] Role of Data & Algorithms in Trustworthy Medical Imaging AI Casey Greene (University of Colorado), [introductory/advanced] Translating Generative AI into Healthcare from Pharmacogenomics to Foundation Models Tina Hernandez-Boussard (Stanford University), [intermediate/advanced] The AI Lifecycle in Healthcare: Evaluation, Deployment, and Responsible Implementation Jianying Hu (IBM Thomas J. Watson Research Center), [intermediate/advanced] Advanced Computing for Biomedical Research and Discovery Jayashree Kalpathy-Cramer (University of Colorado), [introductory] Toward Digital Twins: Multimodal AI for Imaging and Precision Medicine Alex John London (Carnegie Mellon University), [intermediate/advanced] Ethical and Scientific Challenges to Unlocking the Clinical Value of Artificial Intelligence Asoke Nandi (Brunel University of London), [introductory/advanced] Selected Case Studies of Medical Image Segmentation Tom Pollard (Massachusetts Institute of Technology), [introductory/intermediate] Data: The Foundation, Constraint, and Failure Mode of Deployable Health AI Hoifung Poon (Recursion), [advanced] Toward Virtual Patient: AI for Accelerating Medical Discovery Jian Tang (Mila-Qu?bec AI Institute), [introductory/advanced] Generative AI for Protein Design Peter van Ooijen (University of Groningen), [intermediate] Multimodal AI for Adaptive Radiotherapy: Tumor Segmentation, Uncertainty, and Explainability Karin Verspoor (Royal Melbourne Institute of Technology), [introductory/intermediate] AI Scientists and beyond: Roles for GenAI in Biomedical Research and Discovery Jelmer M. Wolterink (University of Twente), [intermediate/advanced] Data Representations in Health Digital Twinning: From Features to Foundation Models HT RESEARCH PERSPECTIVES Human Technopole will organize 3 scientific sessions: Probabilistic Models for (Medical) Image Analysis, by Jan Funke, Florian Jug, Federico Carrara, and Benjamin Salmon AI Models for Computational Biology, by Andrea Sottoriva, Michele Calabr?, and Manuel Dileo AI in Health Data Science, by Francesca Ieva, Andrea Ganna, Michela Carlotta Massi, Nicole Fontana, and Alessia Mapelli SYMPOSIUM A symposium will feature voluntary 10-minute oral presentations on ongoing research projects and industrial developments. Participants interested in presenting should submit a one-page abstract including title, authors, and summary to david at irdta.eu by January 8, 2027. HACKATHON Hands-on team activities will be organized around challenges in generative and multimodal AI for healthcare. The challenges will be released two weeks before the beginning of the school. A jury will evaluate the submissions, and the winners will be announced at the end of February 2027. Winning teams will receive a modest monetary prize, while runners-up will receive certificates of recognition. ORGANIZING COMMITTEE Jan Funke (Milan) Ilaria Guerini (Milan) Francesca Ieva (Milan) Carlos Mart?n-Vide (Tarragona, program chair) Michela Carlotta Massi (Milan, local chair) Santiago Montes (Tarragona, webpage) Sara Morales (Luxembourg, finances) David Silva (London, organization chair) REGISTRATION Registration is available at: https://gemaihc.irdta.eu/2027/registration/ The selection of six courses requested during registration is tentative and non-binding. This information will help estimate demand for logistical planning purposes. As venue capacity is limited, registrations will be processed on a first-come, first-served basis. Registration will close once capacity has been reached. Early registration is strongly recommended. FEES Registration fees include access to all school activities and lunches. Several early registration deadlines are available, and fees vary depending on the registration period. Fees are identical for on-site and online participation. ACCOMMODATION Accommodation suggestions will be provided in due time at: https://gemaihc.irdta.eu/2027/accommodation/ CERTIFICATE Participants will receive a certificate indicating 40 hours of academic activities. This certificate should be suitable for participants seeking ECTS recognition from their home institutions. SPONSORS Companies, institutions, and organizations interested in sponsoring the event may download the sponsorship leaflet from: https://gemaihc.irdta.eu/2027/sponsors/ QUESTIONS AND FURTHER INFORMATION david at irdta.eu ACKNOWLEDGMENTS Human Technopole Universitat Rovira i Virgili IRDTA ? Institute for Research Development, Training and Advice -------------- next part -------------- An HTML attachment was scrubbed... URL: From ASIM.ROY at asu.edu Sat Aug 29 15:40:14 2026 From: ASIM.ROY at asu.edu (Asim Roy) Date: Sat, 29 Aug 2026 19:40:14 +0000 Subject: Connectionists: AI@50 videos...Self-organized learning of objects and their parts In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> Message-ID: Dear Steve, One can certainly claim that ?views? are ?parts.? It will be hard to dispute that. It?s the labeling of these ?parts? that?s the issue. In our Explanation First approach, we make a list of these ?parts? first and then build models based on that list. We have used this approach, for example, to identify different types of ships and aircraft from satellite images. Some standard parts for aircraft are like the fuselage, wings, tail and so on. The Explanation First approach is very similar to an architect understanding the needs of his/her client before designing a house. The ?Explanation First, Models Next? approach is model agnostic. So, we can use one of the ARTMAP models as the underlying model. There is no conflict there. By the way, I explained this Explanation First approach in a simple article in Phalanx last year, the magazine of the Military Operations Research Society (Home - Military Operations Research Society) and it was awarded the best article of 2025. We are starting to use the approach in certain military applications. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen Sent: Saturday, August 29, 2026 6:25 AM To: Asim Roy ; Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen Subject: Re: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Asim, Models of the ARTMAP family (fuzzy ARTMAP, distributed ARTMAP, Gaussian ARTMAP, etc.) are perhaps the simplest kind of self-organizing neural model that can use a teacher (typically the environment, but it can also be a human) to learn to recognize objects, their parts, and their names. The work that my colleagues and I have published has gone far beyond ARTMAP. We have, in a series of articles that can be downloaded from sites.bu.edu/steveg, shown how humans can learn view-, position-, and size-invariant object representations, and use them to carry out actions that achieve valued goals. Our models show how both view-specific and view-invariant representations of an object (say, a particular view of your mother?s face AND an invariant one; namely, that she has a face) can simultaneously guide recognition of mom AND specific actions in response to where the view of her face is looking. The ?views? are ?parts?. I will list some of the relevant article titles and their urls here, for your convenience. Later articles tend to model ever more challenging aspects of self-organizing object recognition and action. DARPA gave us large grants over the years to do this work. Carpenter, G.A., Grossberg, S., and Mehanian, C. (1989). Invariant recognition of cluttered scenes by a self- organizing ART architecture: CORT-X boundary segmentation. Neural Networks, 2, 169-181. https://sites.bu.edu/steveg/files/2016/06/CarGroMeh1989NN.pdf Grossberg, S. and Wyse, L. (1991). Invariant recognition of cluttered scenes by a self-organizing ART architecture: Figure-ground separation. Neural Networks, 1991, 4, 723-742. https://sites.bu.edu/steveg/files/2016/06/GroWyse1991NN.pdf Fazl, A., Grossberg, S., and Mingolla, E. (2009). View-invariant object category learning, recognition, and search: How spatial and object attention are coordinated using surface-based attentional shrouds. Cognitive Psychology, 58, 1-48. https://sites.bu.edu/steveg/files/2016/06/FazGroMin2008.pdf Grossberg, S., and Vladusich, T. (2010). How do children learn to follow gaze, share joint attention, imitate their teachers, and use tools during social interactions? Neural Networks, 23, 940-965. https://sites.bu.edu/steveg/files/2016/06/GrossbergVladusichNN2010.pdf Cao, Y., Grossberg, S., and Markowitz, J. (2011). How does the brain rapidly learn and reorganize view- and positionally-invariant object representations in inferior temporal cortex? Neural Networks, 24, 1050-1061. https://sites.bu.edu/steveg/files/2016/06/NN2853.pdf Grossberg, S., Markowitz, J., and Cao, Y. (2011). On the road to invariant recognition: Explaining tradeoff and morph properties of cells in inferotemporal cortex using multiple-scale task-sensitive attentive learning. Neural Networks, 24, 1036-1049. https://sites.bu.edu/steveg/files/2016/06/GroMarCao2011TR.pdf Grossberg, S., Srinivasan, K., and Yazdabakhsh, A. (2011). On the road to invariant object recognition: How cortical area V2 transforms absolute to relative disparity during 3D vision. Neural Networks, 24, 686-692. https://sites.bu.edu/steveg/files/2016/06/GroSriYaz2011TR.pdf Chang, H.-C., Grossberg, S., and Cao, Y. (2014) Where?s Waldo? How perceptual cognitive, and emotional brain processes cooperate during learning to categorize and find desired objects in a cluttered scene. Frontiers in Integrative Neuroscience, doi: 10.3389/fnint.2014.0043, https://www.frontiersin.org/journals/integrative-neuroscience/articles/10.3389/fnint.2014.00043/full Grossberg, S., Srinivasan, K., and Yazdanbakhsh, A. (2014). Binocular fusion and invariant category learning due to predictive remapping during scanning of a depthful scene with eye movements. Frontiers in Psychology: Perception Science, doi: 10.3389/fpsyg.2014.01457. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.01457/full Grossberg, S. (2019). The embodied brain of SOVEREIGN2: From space-variant conscious percepts during visual search and navigation to learning invariant object categories and cognitive-emotional plans for acquiring valued goals. Frontiers in Computational Neuroscience. Published online: June 25, 2019. https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2019.00036/full Best, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 5:09?AM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Steve, What DARPA seemed to want at that time for computer vision/object recognition is object prediction based on part verification ? if a cat, have you verified its unique parts? That?s the kind of explanation they wanted. ?Teaching? an object recognition system about these parts (to generalize, parts are just subconcepts) is not inconsistent with the way we teach humans. For example, a one- or two-year-old recognizes a car. But we teach them about parts much later - such as what a tire is, a door, steering wheel and so on. How a person?s internal biological model is adjusted from these ?part teachings? I am not sure, but the identification and teaching (labeling) about parts is very much a part of human learning. One might say that LLMs automate this type of learning, but we do provide labels in the descriptions of objects. So, that?s ?teaching.? When we started with multilayered models, the assumption was that at higher levels, there would be part abstractions such as those shown in the figure below - the dog face, legs, tail and so on. The problem was, we couldn?t find these abstractions at higher levels of the network. Even if you found them, we would need to assign labels or names to them ? a dog face, legs, tail and so on. Once you get to assigning these labels, that?s ?teaching.? There is a body of work in computer vision trying to build these abstractions inside CNN models. In our method, we also ?teach? our deep learning models, but in a different way. In general, ?teaching? would be necessary whichever way you do it. We call our approach ?Explanation First, Model Next.? We have reversed the process of model building. Our testing shows gain in overall accuracy compared to some standard object detection models. This DARPA style approach also enables resistance against adversarial attacks. So substantial benefits with the Explanation First approach. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [cid:image003.png at 01DD37B1.AF6B0B30] From: Grossberg, Stephen > Sent: Friday, August 28, 2026 8:01 AM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Asim, Thanks for bringing up the DARPA Explainable AI program. I was invited to submit an article to a Special Issue about Explainable AI that was published when the DARPA program was announced. The editors of the Special Issue knew that Adaptive Resonance Theory, or ART, models automatically discover and incrementally learn over multiple learning trials to pay attention to the predictive, or explainable, critical feature patterns that control successful predictions and are used in learning by the adaptive weights in bottom-up adaptive filters and top-down expectations. ART does this AUTOMATICALLY. You seem to have to use an external teacher to do it. Here is the article: Grossberg, S. (2020). A path towards Explainable AI and autonomous adaptive intelligence: Deep Learning, Adaptive Resonance, and models of perception, emotion, and action. Frontiers in Neurobotics, June 25, 2020. https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2020.00036/full The article?s Abstract illustrates its explanatory range [boldface mine]: * "Biological neural network models whereby brains make minds help to understand autonomous adaptive intelligence. This article summarizes why the dynamics and emergent properties of such models for perception, cognition, emotion, and action are explainable, and thus amenable to being confidently implemented in large-scale applications. Key to their explainability is how these models combine fast activations, or short-term memory (STM) traces, and learned weights, or long-term memory (LTM) traces. Visual and auditory perceptual models have explainable conscious STM representations of visual surfaces and auditory streams in surface-shroud resonances and stream-shroud resonances, respectively. Deep Learning is often used to classify data. However, Deep Learning can experience catastrophic forgetting: At any stage of learning, an unpredictable part of its memory can collapse. Even if it makes some accurate classifications, they are not explainable and thus cannot be used with confidence. Deep Learning shares these problems with the back propagation algorithm, whose computational problems due to non-local weight transport during mismatch learning were described in the 1980s. Deep Learning became popular after very fast computers and huge online databases became available that enabled new applications despite these problems. Adaptive Resonance Theory, or ART, algorithms overcome the computational problems of back propagation and Deep Learning. ART is a self-organizing production system that incrementally learns, using arbitrary combinations of unsupervised and supervised learning and only locally computable quantities, to rapidly classify large non-stationary databases without experiencing catastrophic forgetting. ART classifications and predictions are explainable using the attended critical feature patterns in STM on which they build. The LTM adaptive weights of the fuzzy ARTMAP algorithm induce fuzzy IF-THEN rules that explain what feature combinations predict successful outcomes. ART has been successfully used in multiple large-scale real-world applications, including remote sensing, medical database prediction, and social media data clustering. Also explainable are the MOTIVATOR model of reinforcement learning and cognitive-emotional interactions, and the VITE, DIRECT, DIVA, and SOVEREIGN models for reaching, speech production, spatial navigation, and autonomous adaptive intelligence. These biological models exemplify complementary computing, and use local laws for match learning and mismatch learning that avoid the problems of Deep Learning." Chapter 5 in my 2021 Magnum Opus CONSCIOUS MIND, RESONANT BRAIN: HOW EACH BRAIN MAKES A MIND https://www.amazon.com/dp/0190070552?lv=shuf&channelId=500&plpRedirect=mhFallback provides a self-contained and non-technical overview and synthesis of my results about ART and Explainable AI. The other chapters summarize neural network models of the main processes whereby our brains make our conscious minds in healthy individuals and clinical patients. Best, Steve Stephen Grossberg Wang Professor of Cognitive and Neural Systems Director, Center for Adaptive Systems Emeritus Professor of Mathematics & Statistics, Psychological & Brain Sciences, and Biomedical Engineering Boston University sites.bu.edu/steveg/ steve at bu.edu http://en.wikipedia.org/wiki/Stephen_Grossberg http://scholar.google.com/citations?user=3BIV70wAAAAJ&hl=en https://sites.bu.edu/steveg/files/2021/08/Grossberg-CV-8-14-21.pdf https://youtu.be/9n5AnvFur7I https://www.youtube.com/watch?v=_hBye6JQCh4 https://www.amazon.com/Conscious-Mind-Resonant-Brain-Makes/dp/0190070552 https://www.amazon.com/Your-Creative-Brain-Consciously-Experience/dp/0198965370 From: Connectionists > on behalf of Asim Roy > Date: Friday, August 28, 2026 at 3:51?AM To: Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: AI at 50 videos... Stephen, You raise important issues. There are many ways of dealing with them. Here?s one of the ways. The image below shows what DARPA wanted for Explainable AI. Essentially, find the parts of an object for its prediction. So, for a cat, verify some of its unique parts like the fur, whiskers, and claws. Thus, in our explainable models, we ?teach? an object detection model about these parts. What we get as outputs, for a cat, are predictions for the cat and its parts. There?s a layer of logic that uses these outputs and verifies the existence of the parts before ?finally? predicting that there?s a cat in the image. In essence, it?s neuro-symbolic. One can do the symbolic part with a graphical model too. We can also predict that there?s a cat even if we don?t ?see? all of the parts, such as when we just see its face and not the rest of the body. And, in a similar way, your ?morning coffee? concept can be built from its parts that you mention. There?s a long tradition in computer vision of finding parts of objects inside the model. Our initial conception was that the higher-level filters in a CNN correspond to certain abstractions such as a nose, eyes, ears for a human. That was the distributed representation but based on lower-level abstractions (parts of objects). But years of research showed that one could not find these kinds of lower-level abstractions in standard CNN models. So, what do you do? Well, one way is to force some of the filters to correspond to these parts. There?s a long history to this body of work in computer vision, including that of Hinton, to create abstractions within a model. Both ways, we are ?teaching? the models about ?low-level abstraction.? In a way, in both ways, it?s a hierarchical system. In both ways, we create a neuro-symbolic system. By the way, abstractions are generalizations, they don?t code specific episodic memories. Your ?table,? ?coffee,? ?cup? and all that are abstract notions. Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [A screenshot of a computer AI-generated content may be incorrect.] From: Stephen Jos? Hanson > Sent: Thursday, August 27, 2026 4:26 AM To: Asim Roy >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Asim, I sit here with my coffee neuron and my cup neuron creating a cup of coffee, and my table neuron holding the cup and coffee neurons representing a cup of coffee. On the table, on the floor, noting of course the cup of coffee is not on the floor but on the table. So this is a network of coffee, cups, tables, maybe coded by another neuron in a hierarchical network called morning coffee? Well this is the typical problem with localist model of the brain. How would it work exactly? So far just representing a single episodic event requires hierarchical network of neurons that generalize to what exactly? Do we need many such networks to represent all episodic memories of having a cup of coffee? This is similar to the problem with the socalled "fusiform face area" which with some proper controls doesn't really exist (cf Hanson 2022). What would a small blueberry are in fusiform gyrus be doing with all those faces we recognize? A brain face cloud server? Other accounts show that FG, LO, IN and PFC are at least involved in a network of areas (not neurons) in recognizing a face. But computationally what's the actual underlying functions? The shift to "neurons represent symbols' is not helpful to explain meaning. Remember, we have billions of neurons and 100 trillions connections between them. I'm not saying symbols aren't important and Harnad's Grounding arguments are becoming more critical (and need to be made more explicit) as LLMs start to be embedded in robots. Therefore an arguably simpler task, is to identify "symbols" in LLMs since we have no idea how they work either. Are symbols in their "neurons"? Cheers Stephen On 8/24/26 22:56, Asim Roy wrote: 1. Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? 2. Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? 3. Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195?223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -- [cid:image005.jpg at 01DD37B1.AF6B0B30] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image003.png Type: image/png Size: 333780 bytes Desc: image003.png URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image004.jpg Type: image/jpeg Size: 45946 bytes Desc: image004.jpg URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image005.jpg Type: image/jpeg Size: 57383 bytes Desc: image005.jpg URL: From andypotatohy at gmail.com Sat Aug 29 17:38:48 2026 From: andypotatohy at gmail.com (Yu (Andy) Huang) Date: Sat, 29 Aug 2026 15:38:48 -0600 Subject: Connectionists: =?utf-8?q?PhD_Student_Position_available_Spring/F?= =?utf-8?q?all_2027_=E2=80=94_AI_in_Biomedical_Engineering?= Message-ID: PhD Student Position available Spring/Fall 2027 ? AI in Biomedical Engineering H-Lab | Department of Computer Science | University of Colorado at Colorado Springs (UCCS) PI: Yu (Andy) Huang, Ph.D. ? Assistant Professor, Department of Computer Science, UCCS Contact: *yhuang6 at uccs.edu* *About the Lab* The newly established H-Lab, directed by Dr. Huang, sits at the boundary between physics, engineering, computer science, and biomedicine. Dr. Huang was trained as a biomedical engineer and created ROAST, the field-standard open-source pipeline for personalized modeling of transcranial electric stimulation (TES), spending five years helping commercialize it in industry before returning to academia. He has also built machine-learning systems, developed with practicing radiologists, that detect disease from medical images at expert-level accuracy. The H-Lab?s current interests focus on developing AI systems for biomedical engineering and bioinformatics. Future interests include understanding modern deep-learning systems, embodied AI, neuromorphic computing, and artificial life. For the lab?s longer-term research visions, please refer to the *PI?s webpage* . *Research directions* We are recruiting one PhD student to help build AI systems for biomedical engineering & informatics. Specific research topics include but not limited to: - Algorithms development - Image processing, computer vision - Computational photography - Finite element modeling - Machine learning, deep learning - Large language models - Open-source software development - Synergistic activities from topics above with AI, such as agentic AI, AI for scientists, and AI for engineers - Philosophy of AI The successful candidate will have access to deep-learning machines and high-performance computing clusters at UCCS with hundreds of GPUs. The successful candidate will also have opportunities to collaborate with the medical device industry and prestigious medical centers. If any of this excites you, we want to hear from you. *What you must have:* - Bachelor?s and/or Master's degree in computer science, electrical engineering, biomedical engineering, physics, applied mathematics, or related areas. - Good programming knowledge and experience with Python and/or MATLAB. - Fluent written and spoken English. - Enthusiasm for cutting-edge research, team spirit, and capability of independent problem-solving. *It?s nice to also have:* - Interest, prior knowledge and experience in one or more of the following: machine learning, deep learning, computational neuroscience, biomedical signal/image processing, finite element modeling, scientific computing, large language models, agentic AI, and philosophy of AI - Prior research experience (thesis, publications, open-source contributions) *How to Apply* The position will open in Spring 2027 or Fall 2027 semester, fully funded with stipend and tuition waiver for 2 years. Interested applicants should apply via *the department website* *by October 1st* for the Spring admission or *by February 1st* for the Fall admission. Applicants should note in their Personal Statement on *which research direction(s) above interest you most and why*, and *mention Dr. Huang as your doctoral advisor*. Applicants are also encouraged to send their CV and unofficial transcript in PDF format to Dr. Huang at *yhuang6 at uccs.edu* . *Why UCCS Computer Science* One of the few departments nationally offering both a PhD in Computer Science and a standalone PhD in Cybersecurity ? the latter designated an NSA National Center of Academic Excellence in Cyber Research (CAE-R), the first such designation in Colorado. Faculty expertise spanning AI, machine learning, computer vision, NLP, biomedical informatics, cybersecurity, computer networks, software engineering, and multiagent systems, across 10 cross-disciplinary research labs. Research funded by NSF, DoD, and AFOSR, including four NSF CAREER awards and an AFOSR Young Investigator award among current faculty. Located in Colorado Springs, with deep aerospace, defense, and tech industry ties (alumni at Google, IBM, Lockheed Martin, MITRE, Northrop Grumman, and beyond) and a fast-growing research profile. -- *Yu (Andy) Huang, Ph.D.* Tenure-track Assistant Professor of Computer Science Department of Computer Science University of Colorado at Colorado Springs 1420 Austin Bluffs Pkway, Colorado Springs, CO 80918 Email: yhuang6 at uccs.edu, andypotatohy at gmail.com Website -------------- next part -------------- An HTML attachment was scrubbed... URL: From ASIM.ROY at asu.edu Sat Aug 29 20:31:54 2026 From: ASIM.ROY at asu.edu (Asim Roy) Date: Sun, 30 Aug 2026 00:31:54 +0000 Subject: Connectionists: AI@50 videos...Self-organized learning of objects and their parts In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> Message-ID: Dear Steve, I am fully aware that the parts-based explanation has limits. For example, you can?t define the parts of a rock or a boulder easily, if at all, unless you bring in texture, color, and all of that. Part-based explanation is based on human understanding of objects in terms of parts and is consistent with ?teaching? humans about parts of objects. The general idea is to engage with model users and understand which parts they want to verify if a parts-based approach is acceptable for explanation. From our experience, one of the advantages of the part-based approach is that it can in fact deal with occlusions, camouflage, noise, and other kinds of distortions. For example, with this approach, if you just see the face of a cat and the body is occluded (behind a door), you can still recognize the object as a cat and justify with the parts-based logic. And the model can do that even though it was never trained on such images of cats. One can think of the parts-based approach extending the concept of invariance. We can deal with variety. For example, small drones come in different shapes and sizes, and we can do part-based recognition of flying drones. I think we all are aware of your wide body of work, which is extensive. However, I look more closely at the computer vision literature, and they directly address the problem of recognizing parts of objects within their models. Thus, my approach aligns with that body of work. And it?s consistent with the way humans learn and explain. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen Sent: Saturday, August 29, 2026 4:24 PM To: Asim Roy ; Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen Subject: Re: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Asim, You write: "make a list of these ?parts? first and then build models based on that list.? This approach was shown decades ago to be unable to deal with novel shapes and their parts. Irv Biederman is perhaps the most distinguished psychologist to have promoted it; e.g., in his 1987 Psychological Review article (94, 115-147) entitled: Recognition-by-Components: A Theory of Human Image Understanding His approach was shown to have only limited explanatory power, since the variety of POSSIBLE novel two-dimensional and three-dimensional parts and wholes exceeds the explanatory range of the kinds of ?models? that you use. Natural objects in the real world cannot easily, if at all, be described by your approach. The speed with which you replied to my extensive list of article titles and abstracts leads me to think that you did not have enough time to even scan them. They offer rigorous computational solutions to the problem that you have posed. Please take the time to read at least the article titles and abstracts. Your approach says nothing about how a two-dimensional picture of a shape is consciously seen as the three-dimensional representation of the shape. The Necker Cube is one of hundreds of examples. Your approach also says nothing about how perceptual grouping binds distributed features into parts or wholes, depending upon Gestalt grouping laws, or how figure-ground separation occurs; notably, how we consciously see the unoccluded parts of a partly occluded object, but often only recognize, without seeing, the occluded parts. And so on... Here are a few additional relevant articles: Cohen, M.A. and Grossberg, S. (1990). Unitized recognition codes for parts and wholes: The unique cue in configural discriminations. In M. Commons, R. Herrnstein, S. Kosslyn, and D. Mumford (Eds.), Computational and clinical approaches to pattern recognition and concept formation. Hillsdale, NJ: Erlbaum. https://sites.bu.edu/steveg/files/2016/06/CohenGro1990UniRecogCodes.pdf Grossberg, S. (2016). Cortical dynamics of figure-ground separation in response to 2D pictures and 3D scenes: How V2 combines border ownership, stereoscopic cues, and Gestalt grouping rules. Frontiers in Psychology. 26 January 2016. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2015.02054/full Grossberg, S., and Swaminathan, G. (2004). A laminar cortical model for 3D perception of slanted and curved surfaces and of 2D images: development, attention and bistability. Vision Research, 44, 1147-1187. https://sites.bu.edu/steveg/files/2016/06/GroSwa2004VR.pdf Congratulations on winning the best article of 2025 award in the magazine of the Military Operations Research Society! Our models have been deployed in software and hardware by MIT Lincoln Lab and other National Labs. We were awarded millions of dollars of competitive grants and contracts by essentially all the major funding agencies in Washington and elsewhere, including ARO, AFOSR, ONR, DARPA, NSF, NIH, Hughes Research Lab, etc. to develop them. We won millions of dollars in grants and contracts because our models work in the real world. Best again, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 3:40?PM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Steve, One can certainly claim that ?views? are ?parts.? It will be hard to dispute that. It?s the labeling of these ?parts? that?s the issue. In our Explanation First approach, we make a list of these ?parts? first and then build models based on that list. We have used this approach, for example, to identify different types of ships and aircraft from satellite images. Some standard parts for aircraft are like the fuselage, wings, tail and so on. The Explanation First approach is very similar to an architect understanding the needs of his/her client before designing a house. The ?Explanation First, Models Next? approach is model agnostic. So, we can use one of the ARTMAP models as the underlying model. There is no conflict there. By the way, I explained this Explanation First approach in a simple article in Phalanx last year, the magazine of the Military Operations Research Society (Home - Military Operations Research Society) and it was awarded the best article of 2025. We are starting to use the approach in certain military applications. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen > Sent: Saturday, August 29, 2026 6:25 AM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Asim, Models of the ARTMAP family (fuzzy ARTMAP, distributed ARTMAP, Gaussian ARTMAP, etc.) are perhaps the simplest kind of self-organizing neural model that can use a teacher (typically the environment, but it can also be a human) to learn to recognize objects, their parts, and their names. The work that my colleagues and I have published has gone far beyond ARTMAP. We have, in a series of articles that can be downloaded from sites.bu.edu/steveg, shown how humans can learn view-, position-, and size-invariant object representations, and use them to carry out actions that achieve valued goals. Our models show how both view-specific and view-invariant representations of an object (say, a particular view of your mother?s face AND an invariant one; namely, that she has a face) can simultaneously guide recognition of mom AND specific actions in response to where the view of her face is looking. The ?views? are ?parts?. I will list some of the relevant article titles and their urls here, for your convenience. Later articles tend to model ever more challenging aspects of self-organizing object recognition and action. DARPA gave us large grants over the years to do this work. Carpenter, G.A., Grossberg, S., and Mehanian, C. (1989). Invariant recognition of cluttered scenes by a self- organizing ART architecture: CORT-X boundary segmentation. Neural Networks, 2, 169-181. https://sites.bu.edu/steveg/files/2016/06/CarGroMeh1989NN.pdf Grossberg, S. and Wyse, L. (1991). Invariant recognition of cluttered scenes by a self-organizing ART architecture: Figure-ground separation. Neural Networks, 1991, 4, 723-742. https://sites.bu.edu/steveg/files/2016/06/GroWyse1991NN.pdf Fazl, A., Grossberg, S., and Mingolla, E. (2009). View-invariant object category learning, recognition, and search: How spatial and object attention are coordinated using surface-based attentional shrouds. Cognitive Psychology, 58, 1-48. https://sites.bu.edu/steveg/files/2016/06/FazGroMin2008.pdf Grossberg, S., and Vladusich, T. (2010). How do children learn to follow gaze, share joint attention, imitate their teachers, and use tools during social interactions? Neural Networks, 23, 940-965. https://sites.bu.edu/steveg/files/2016/06/GrossbergVladusichNN2010.pdf Cao, Y., Grossberg, S., and Markowitz, J. (2011). How does the brain rapidly learn and reorganize view- and positionally-invariant object representations in inferior temporal cortex? Neural Networks, 24, 1050-1061. https://sites.bu.edu/steveg/files/2016/06/NN2853.pdf Grossberg, S., Markowitz, J., and Cao, Y. (2011). On the road to invariant recognition: Explaining tradeoff and morph properties of cells in inferotemporal cortex using multiple-scale task-sensitive attentive learning. Neural Networks, 24, 1036-1049. https://sites.bu.edu/steveg/files/2016/06/GroMarCao2011TR.pdf Grossberg, S., Srinivasan, K., and Yazdabakhsh, A. (2011). On the road to invariant object recognition: How cortical area V2 transforms absolute to relative disparity during 3D vision. Neural Networks, 24, 686-692. https://sites.bu.edu/steveg/files/2016/06/GroSriYaz2011TR.pdf Chang, H.-C., Grossberg, S., and Cao, Y. (2014) Where?s Waldo? How perceptual cognitive, and emotional brain processes cooperate during learning to categorize and find desired objects in a cluttered scene. Frontiers in Integrative Neuroscience, doi: 10.3389/fnint.2014.0043, https://www.frontiersin.org/journals/integrative-neuroscience/articles/10.3389/fnint.2014.00043/full Grossberg, S., Srinivasan, K., and Yazdanbakhsh, A. (2014). Binocular fusion and invariant category learning due to predictive remapping during scanning of a depthful scene with eye movements. Frontiers in Psychology: Perception Science, doi: 10.3389/fpsyg.2014.01457. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.01457/full Grossberg, S. (2019). The embodied brain of SOVEREIGN2: From space-variant conscious percepts during visual search and navigation to learning invariant object categories and cognitive-emotional plans for acquiring valued goals. Frontiers in Computational Neuroscience. Published online: June 25, 2019. https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2019.00036/full Best, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 5:09?AM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Steve, What DARPA seemed to want at that time for computer vision/object recognition is object prediction based on part verification ? if a cat, have you verified its unique parts? That?s the kind of explanation they wanted. ?Teaching? an object recognition system about these parts (to generalize, parts are just subconcepts) is not inconsistent with the way we teach humans. For example, a one- or two-year-old recognizes a car. But we teach them about parts much later - such as what a tire is, a door, steering wheel and so on. How a person?s internal biological model is adjusted from these ?part teachings? I am not sure, but the identification and teaching (labeling) about parts is very much a part of human learning. One might say that LLMs automate this type of learning, but we do provide labels in the descriptions of objects. So, that?s ?teaching.? When we started with multilayered models, the assumption was that at higher levels, there would be part abstractions such as those shown in the figure below - the dog face, legs, tail and so on. The problem was, we couldn?t find these abstractions at higher levels of the network. Even if you found them, we would need to assign labels or names to them ? a dog face, legs, tail and so on. Once you get to assigning these labels, that?s ?teaching.? There is a body of work in computer vision trying to build these abstractions inside CNN models. In our method, we also ?teach? our deep learning models, but in a different way. In general, ?teaching? would be necessary whichever way you do it. We call our approach ?Explanation First, Model Next.? We have reversed the process of model building. Our testing shows gain in overall accuracy compared to some standard object detection models. This DARPA style approach also enables resistance against adversarial attacks. So substantial benefits with the Explanation First approach. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [cid:image001.png at 01DD37D5.01A7D180] From: Grossberg, Stephen > Sent: Friday, August 28, 2026 8:01 AM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Asim, Thanks for bringing up the DARPA Explainable AI program. I was invited to submit an article to a Special Issue about Explainable AI that was published when the DARPA program was announced. The editors of the Special Issue knew that Adaptive Resonance Theory, or ART, models automatically discover and incrementally learn over multiple learning trials to pay attention to the predictive, or explainable, critical feature patterns that control successful predictions and are used in learning by the adaptive weights in bottom-up adaptive filters and top-down expectations. ART does this AUTOMATICALLY. You seem to have to use an external teacher to do it. Here is the article: Grossberg, S. (2020). A path towards Explainable AI and autonomous adaptive intelligence: Deep Learning, Adaptive Resonance, and models of perception, emotion, and action. Frontiers in Neurobotics, June 25, 2020. https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2020.00036/full The article?s Abstract illustrates its explanatory range [boldface mine]: * "Biological neural network models whereby brains make minds help to understand autonomous adaptive intelligence. This article summarizes why the dynamics and emergent properties of such models for perception, cognition, emotion, and action are explainable, and thus amenable to being confidently implemented in large-scale applications. Key to their explainability is how these models combine fast activations, or short-term memory (STM) traces, and learned weights, or long-term memory (LTM) traces. Visual and auditory perceptual models have explainable conscious STM representations of visual surfaces and auditory streams in surface-shroud resonances and stream-shroud resonances, respectively. Deep Learning is often used to classify data. However, Deep Learning can experience catastrophic forgetting: At any stage of learning, an unpredictable part of its memory can collapse. Even if it makes some accurate classifications, they are not explainable and thus cannot be used with confidence. Deep Learning shares these problems with the back propagation algorithm, whose computational problems due to non-local weight transport during mismatch learning were described in the 1980s. Deep Learning became popular after very fast computers and huge online databases became available that enabled new applications despite these problems. Adaptive Resonance Theory, or ART, algorithms overcome the computational problems of back propagation and Deep Learning. ART is a self-organizing production system that incrementally learns, using arbitrary combinations of unsupervised and supervised learning and only locally computable quantities, to rapidly classify large non-stationary databases without experiencing catastrophic forgetting. ART classifications and predictions are explainable using the attended critical feature patterns in STM on which they build. The LTM adaptive weights of the fuzzy ARTMAP algorithm induce fuzzy IF-THEN rules that explain what feature combinations predict successful outcomes. ART has been successfully used in multiple large-scale real-world applications, including remote sensing, medical database prediction, and social media data clustering. Also explainable are the MOTIVATOR model of reinforcement learning and cognitive-emotional interactions, and the VITE, DIRECT, DIVA, and SOVEREIGN models for reaching, speech production, spatial navigation, and autonomous adaptive intelligence. These biological models exemplify complementary computing, and use local laws for match learning and mismatch learning that avoid the problems of Deep Learning." Chapter 5 in my 2021 Magnum Opus CONSCIOUS MIND, RESONANT BRAIN: HOW EACH BRAIN MAKES A MIND https://www.amazon.com/dp/0190070552?lv=shuf&channelId=500&plpRedirect=mhFallback provides a self-contained and non-technical overview and synthesis of my results about ART and Explainable AI. The other chapters summarize neural network models of the main processes whereby our brains make our conscious minds in healthy individuals and clinical patients. Best, Steve Stephen Grossberg Wang Professor of Cognitive and Neural Systems Director, Center for Adaptive Systems Emeritus Professor of Mathematics & Statistics, Psychological & Brain Sciences, and Biomedical Engineering Boston University sites.bu.edu/steveg/ steve at bu.edu http://en.wikipedia.org/wiki/Stephen_Grossberg http://scholar.google.com/citations?user=3BIV70wAAAAJ&hl=en https://sites.bu.edu/steveg/files/2021/08/Grossberg-CV-8-14-21.pdf https://youtu.be/9n5AnvFur7I https://www.youtube.com/watch?v=_hBye6JQCh4 https://www.amazon.com/Conscious-Mind-Resonant-Brain-Makes/dp/0190070552 https://www.amazon.com/Your-Creative-Brain-Consciously-Experience/dp/0198965370 From: Connectionists > on behalf of Asim Roy > Date: Friday, August 28, 2026 at 3:51?AM To: Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: AI at 50 videos... Stephen, You raise important issues. There are many ways of dealing with them. Here?s one of the ways. The image below shows what DARPA wanted for Explainable AI. Essentially, find the parts of an object for its prediction. So, for a cat, verify some of its unique parts like the fur, whiskers, and claws. Thus, in our explainable models, we ?teach? an object detection model about these parts. What we get as outputs, for a cat, are predictions for the cat and its parts. There?s a layer of logic that uses these outputs and verifies the existence of the parts before ?finally? predicting that there?s a cat in the image. In essence, it?s neuro-symbolic. One can do the symbolic part with a graphical model too. We can also predict that there?s a cat even if we don?t ?see? all of the parts, such as when we just see its face and not the rest of the body. And, in a similar way, your ?morning coffee? concept can be built from its parts that you mention. There?s a long tradition in computer vision of finding parts of objects inside the model. Our initial conception was that the higher-level filters in a CNN correspond to certain abstractions such as a nose, eyes, ears for a human. That was the distributed representation but based on lower-level abstractions (parts of objects). But years of research showed that one could not find these kinds of lower-level abstractions in standard CNN models. So, what do you do? Well, one way is to force some of the filters to correspond to these parts. There?s a long history to this body of work in computer vision, including that of Hinton, to create abstractions within a model. Both ways, we are ?teaching? the models about ?low-level abstraction.? In a way, in both ways, it?s a hierarchical system. In both ways, we create a neuro-symbolic system. By the way, abstractions are generalizations, they don?t code specific episodic memories. Your ?table,? ?coffee,? ?cup? and all that are abstract notions. Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [A screenshot of a computer AI-generated content may be incorrect.] From: Stephen Jos? Hanson > Sent: Thursday, August 27, 2026 4:26 AM To: Asim Roy >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Asim, I sit here with my coffee neuron and my cup neuron creating a cup of coffee, and my table neuron holding the cup and coffee neurons representing a cup of coffee. On the table, on the floor, noting of course the cup of coffee is not on the floor but on the table. So this is a network of coffee, cups, tables, maybe coded by another neuron in a hierarchical network called morning coffee? Well this is the typical problem with localist model of the brain. How would it work exactly? So far just representing a single episodic event requires hierarchical network of neurons that generalize to what exactly? Do we need many such networks to represent all episodic memories of having a cup of coffee? This is similar to the problem with the socalled "fusiform face area" which with some proper controls doesn't really exist (cf Hanson 2022). What would a small blueberry are in fusiform gyrus be doing with all those faces we recognize? A brain face cloud server? Other accounts show that FG, LO, IN and PFC are at least involved in a network of areas (not neurons) in recognizing a face. But computationally what's the actual underlying functions? The shift to "neurons represent symbols' is not helpful to explain meaning. Remember, we have billions of neurons and 100 trillions connections between them. I'm not saying symbols aren't important and Harnad's Grounding arguments are becoming more critical (and need to be made more explicit) as LLMs start to be embedded in robots. Therefore an arguably simpler task, is to identify "symbols" in LLMs since we have no idea how they work either. Are symbols in their "neurons"? Cheers Stephen On 8/24/26 22:56, Asim Roy wrote: 1. Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? 2. Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? 3. Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195?223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -- [cid:image003.jpg at 01DD37D5.01A7D180] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image001.png Type: image/png Size: 333780 bytes Desc: image001.png URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image002.jpg Type: image/jpeg Size: 45946 bytes Desc: image002.jpg URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image003.jpg Type: image/jpeg Size: 57383 bytes Desc: image003.jpg URL: From ASIM.ROY at asu.edu Sun Aug 30 18:36:08 2026 From: ASIM.ROY at asu.edu (Asim Roy) Date: Sun, 30 Aug 2026 22:36:08 +0000 Subject: Connectionists: AI@50 videos...Specialized applications vs. general-purpose self-organizing neural networks for autonomous adaptive intelligence In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> Message-ID: Dear Steve, Thanks. Appreciate it. Best, Asim From: Grossberg, Stephen Sent: Sunday, August 30, 2026 3:25 PM To: Asim Roy ; Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos...Specialized applications vs. general-purpose self-organizing neural networks for autonomous adaptive intelligence Dear Asim, Your reply is completely consistent with my previous remarks. Keep up the good work! Best, Steve Get Outlook for iOS ________________________________ From: Asim Roy > Sent: Sunday, 30 August 2026 18:13:10 To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Specialized applications vs. general-purpose self-organizing neural networks for autonomous adaptive intelligence Dear Steve, As I mentioned earlier, the approach is neurosymbolic. We ?teach? a neuro model about the ?symbols? we are interested in. It?s as simple as that. And we then process the symbolic output in the symbol processing unit. The symbols might be cat and its parts. Or airplanes and ships and their parts. Or drones and their parts. It?s fairly general idea and we can tie it down to how humans are taught to recognize objects. We have not tried the approach in other domains, but, in theory, it should be applicable. We do break down speech and sound into parts. And sentences into words and then syllables. Part-based recognition is powerful. It can handle adversarial attacks quite well. I would consider adversarial attacks as an extension of the concept of invariance. Our approach is model agnostic in the sense that we can as well use your ART family of models as the ?neuro? engine. Right now, we simply use fairly standard off-the-shelf object detection models. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen > Sent: Sunday, August 30, 2026 7:37 AM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Specialized applications vs. general-purpose self-organizing neural networks for autonomous adaptive intelligence Dear Asim, I totally agree that, if you have developed an application that works, use it! Over the years, I and my colleagues have done the same. We have regularly abstracted from our biological neural network models to design fast algorithms for efficient operation in specific application domains. E.g., Srinivasa, N., Bhattacharyya, R., Sundareswara, R., Lee, C., and Grossberg, S. (2012) A bio-inspired kinematic controller for obstacle avoidance during reaching tasks with real robots. Neural Networks, 35, 54-69. https://sites.bu.edu/steveg/files/2016/06/NN_Final.pdf Ivey, R., Bullock, D., and Grossberg, S. (2011). A neuromorphic model of spatial lookahead planning. Neural Networks, 24, 257-266. https://sites.bu.edu/steveg/files/2016/06/IveBulGro2011.pdf Grossberg, S., & Huang, T.-R. (2009). ARTSCENE: A neural system for natural scene classification. Journal of Vision, 9(4):6, 1-19. https://jov.arvojournals.org/article.aspx?articleid=2193487 Hong, S. and Grossberg, S. (2004). A neuromorphic model for achromatic and chromatic surface representation of natural images. Neural Networks, 2004, 17, 787-808. https://sites.bu.edu/steveg/files/2016/06/HongGro2004NN.pdf There are at least two possible problems with such applications: They tend to be domain-specific and, if you try to use them in a different task environment, you often hit a Brick Wall. In my modeling work during the past 69 years of how our brains make our conscious minds, I have never hit a Brick Wall. In fact, earlier principles, mechanisms, and architectures have always formed a foundation for the next discoveries. Indeed, ALL of my models can be expressed in terms of specializations of a few basic equations, and a somewhat larger number of modules or microcircuits, that are assembled into what I call MODAL architectures (for different MODALities), such as vision and image processing; audition, speech, language, and music; attention, categorization, cognitive information processing, and prediction; motivation and cognitive-emotional interactions; adaptive sensory-motor control, including reaching, standing, walking, running, tool use, dancing; and so on. That is why I am still discovering and developing new models, as my web page sites.bu.edu/steveg illustrates. It is also possible, if your current implementations do not achieve all their goals, or the goals change, for you to turn to our models for concepts and mechanisms that you can specialize for your own applications. More generally, taken together, the almost 600 downloadable archival articles on my web page provide a blueprint for the revolutionary computational paradigm of autonomous adaptive intelligence. Self-driving cars are just one example that is currently being developed. In the next century, particularly as societies and technologies need to rapidly adapt to new economic, political, and environmental challenges, increasing autonomous, adaptively intelligent solutions will help us to survive, or at least that is my hope. Best, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 8:32?PM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Steve, I am fully aware that the parts-based explanation has limits. For example, you can?t define the parts of a rock or a boulder easily, if at all, unless you bring in texture, color, and all of that. Part-based explanation is based on human understanding of objects in terms of parts and is consistent with ?teaching? humans about parts of objects. The general idea is to engage with model users and understand which parts they want to verify if a parts-based approach is acceptable for explanation. From our experience, one of the advantages of the part-based approach is that it can in fact deal with occlusions, camouflage, noise, and other kinds of distortions. For example, with this approach, if you just see the face of a cat and the body is occluded (behind a door), you can still recognize the object as a cat and justify with the parts-based logic. And the model can do that even though it was never trained on such images of cats. One can think of the parts-based approach extending the concept of invariance. We can deal with variety. For example, small drones come in different shapes and sizes, and we can do part-based recognition of flying drones. I think we all are aware of your wide body of work, which is extensive. However, I look more closely at the computer vision literature, and they directly address the problem of recognizing parts of objects within their models. Thus, my approach aligns with that body of work. And it?s consistent with the way humans learn and explain. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen > Sent: Saturday, August 29, 2026 4:24 PM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Asim, You write: "make a list of these ?parts? first and then build models based on that list.? This approach was shown decades ago to be unable to deal with novel shapes and their parts. Irv Biederman is perhaps the most distinguished psychologist to have promoted it; e.g., in his 1987 Psychological Review article (94, 115-147) entitled: Recognition-by-Components: A Theory of Human Image Understanding His approach was shown to have only limited explanatory power, since the variety of POSSIBLE novel two-dimensional and three-dimensional parts and wholes exceeds the explanatory range of the kinds of ?models? that you use. Natural objects in the real world cannot easily, if at all, be described by your approach. The speed with which you replied to my extensive list of article titles and abstracts leads me to think that you did not have enough time to even scan them. They offer rigorous computational solutions to the problem that you have posed. Please take the time to read at least the article titles and abstracts. Your approach says nothing about how a two-dimensional picture of a shape is consciously seen as the three-dimensional representation of the shape. The Necker Cube is one of hundreds of examples. Your approach also says nothing about how perceptual grouping binds distributed features into parts or wholes, depending upon Gestalt grouping laws, or how figure-ground separation occurs; notably, how we consciously see the unoccluded parts of a partly occluded object, but often only recognize, without seeing, the occluded parts. And so on... Here are a few additional relevant articles: Cohen, M.A. and Grossberg, S. (1990). Unitized recognition codes for parts and wholes: The unique cue in configural discriminations. In M. Commons, R. Herrnstein, S. Kosslyn, and D. Mumford (Eds.), Computational and clinical approaches to pattern recognition and concept formation. Hillsdale, NJ: Erlbaum. https://sites.bu.edu/steveg/files/2016/06/CohenGro1990UniRecogCodes.pdf Grossberg, S. (2016). Cortical dynamics of figure-ground separation in response to 2D pictures and 3D scenes: How V2 combines border ownership, stereoscopic cues, and Gestalt grouping rules. Frontiers in Psychology. 26 January 2016. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2015.02054/full Grossberg, S., and Swaminathan, G. (2004). A laminar cortical model for 3D perception of slanted and curved surfaces and of 2D images: development, attention and bistability. Vision Research, 44, 1147-1187. https://sites.bu.edu/steveg/files/2016/06/GroSwa2004VR.pdf Congratulations on winning the best article of 2025 award in the magazine of the Military Operations Research Society! Our models have been deployed in software and hardware by MIT Lincoln Lab and other National Labs. We were awarded millions of dollars of competitive grants and contracts by essentially all the major funding agencies in Washington and elsewhere, including ARO, AFOSR, ONR, DARPA, NSF, NIH, Hughes Research Lab, etc. to develop them. We won millions of dollars in grants and contracts because our models work in the real world. Best again, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 3:40?PM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Steve, One can certainly claim that ?views? are ?parts.? It will be hard to dispute that. It?s the labeling of these ?parts? that?s the issue. In our Explanation First approach, we make a list of these ?parts? first and then build models based on that list. We have used this approach, for example, to identify different types of ships and aircraft from satellite images. Some standard parts for aircraft are like the fuselage, wings, tail and so on. The Explanation First approach is very similar to an architect understanding the needs of his/her client before designing a house. The ?Explanation First, Models Next? approach is model agnostic. So, we can use one of the ARTMAP models as the underlying model. There is no conflict there. By the way, I explained this Explanation First approach in a simple article in Phalanx last year, the magazine of the Military Operations Research Society (Home - Military Operations Research Society) and it was awarded the best article of 2025. We are starting to use the approach in certain military applications. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen > Sent: Saturday, August 29, 2026 6:25 AM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Asim, Models of the ARTMAP family (fuzzy ARTMAP, distributed ARTMAP, Gaussian ARTMAP, etc.) are perhaps the simplest kind of self-organizing neural model that can use a teacher (typically the environment, but it can also be a human) to learn to recognize objects, their parts, and their names. The work that my colleagues and I have published has gone far beyond ARTMAP. We have, in a series of articles that can be downloaded from sites.bu.edu/steveg, shown how humans can learn view-, position-, and size-invariant object representations, and use them to carry out actions that achieve valued goals. Our models show how both view-specific and view-invariant representations of an object (say, a particular view of your mother?s face AND an invariant one; namely, that she has a face) can simultaneously guide recognition of mom AND specific actions in response to where the view of her face is looking. The ?views? are ?parts?. I will list some of the relevant article titles and their urls here, for your convenience. Later articles tend to model ever more challenging aspects of self-organizing object recognition and action. DARPA gave us large grants over the years to do this work. Carpenter, G.A., Grossberg, S., and Mehanian, C. (1989). Invariant recognition of cluttered scenes by a self- organizing ART architecture: CORT-X boundary segmentation. Neural Networks, 2, 169-181. https://sites.bu.edu/steveg/files/2016/06/CarGroMeh1989NN.pdf Grossberg, S. and Wyse, L. (1991). Invariant recognition of cluttered scenes by a self-organizing ART architecture: Figure-ground separation. Neural Networks, 1991, 4, 723-742. https://sites.bu.edu/steveg/files/2016/06/GroWyse1991NN.pdf Fazl, A., Grossberg, S., and Mingolla, E. (2009). View-invariant object category learning, recognition, and search: How spatial and object attention are coordinated using surface-based attentional shrouds. Cognitive Psychology, 58, 1-48. https://sites.bu.edu/steveg/files/2016/06/FazGroMin2008.pdf Grossberg, S., and Vladusich, T. (2010). How do children learn to follow gaze, share joint attention, imitate their teachers, and use tools during social interactions? Neural Networks, 23, 940-965. https://sites.bu.edu/steveg/files/2016/06/GrossbergVladusichNN2010.pdf Cao, Y., Grossberg, S., and Markowitz, J. (2011). How does the brain rapidly learn and reorganize view- and positionally-invariant object representations in inferior temporal cortex? Neural Networks, 24, 1050-1061. https://sites.bu.edu/steveg/files/2016/06/NN2853.pdf Grossberg, S., Markowitz, J., and Cao, Y. (2011). On the road to invariant recognition: Explaining tradeoff and morph properties of cells in inferotemporal cortex using multiple-scale task-sensitive attentive learning. Neural Networks, 24, 1036-1049. https://sites.bu.edu/steveg/files/2016/06/GroMarCao2011TR.pdf Grossberg, S., Srinivasan, K., and Yazdabakhsh, A. (2011). On the road to invariant object recognition: How cortical area V2 transforms absolute to relative disparity during 3D vision. Neural Networks, 24, 686-692. https://sites.bu.edu/steveg/files/2016/06/GroSriYaz2011TR.pdf Chang, H.-C., Grossberg, S., and Cao, Y. (2014) Where?s Waldo? How perceptual cognitive, and emotional brain processes cooperate during learning to categorize and find desired objects in a cluttered scene. Frontiers in Integrative Neuroscience, doi: 10.3389/fnint.2014.0043, https://www.frontiersin.org/journals/integrative-neuroscience/articles/10.3389/fnint.2014.00043/full Grossberg, S., Srinivasan, K., and Yazdanbakhsh, A. (2014). Binocular fusion and invariant category learning due to predictive remapping during scanning of a depthful scene with eye movements. Frontiers in Psychology: Perception Science, doi: 10.3389/fpsyg.2014.01457. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.01457/full Grossberg, S. (2019). The embodied brain of SOVEREIGN2: From space-variant conscious percepts during visual search and navigation to learning invariant object categories and cognitive-emotional plans for acquiring valued goals. Frontiers in Computational Neuroscience. Published online: June 25, 2019. https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2019.00036/full Best, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 5:09?AM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Steve, What DARPA seemed to want at that time for computer vision/object recognition is object prediction based on part verification ? if a cat, have you verified its unique parts? That?s the kind of explanation they wanted. ?Teaching? an object recognition system about these parts (to generalize, parts are just subconcepts) is not inconsistent with the way we teach humans. For example, a one- or two-year-old recognizes a car. But we teach them about parts much later - such as what a tire is, a door, steering wheel and so on. How a person?s internal biological model is adjusted from these ?part teachings? I am not sure, but the identification and teaching (labeling) about parts is very much a part of human learning. One might say that LLMs automate this type of learning, but we do provide labels in the descriptions of objects. So, that?s ?teaching.? When we started with multilayered models, the assumption was that at higher levels, there would be part abstractions such as those shown in the figure below - the dog face, legs, tail and so on. The problem was, we couldn?t find these abstractions at higher levels of the network. Even if you found them, we would need to assign labels or names to them ? a dog face, legs, tail and so on. Once you get to assigning these labels, that?s ?teaching.? There is a body of work in computer vision trying to build these abstractions inside CNN models. In our method, we also ?teach? our deep learning models, but in a different way. In general, ?teaching? would be necessary whichever way you do it. We call our approach ?Explanation First, Model Next.? We have reversed the process of model building. Our testing shows gain in overall accuracy compared to some standard object detection models. This DARPA style approach also enables resistance against adversarial attacks. So substantial benefits with the Explanation First approach. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [cid:image001.png at 01DD3895.457E71A0] From: Grossberg, Stephen > Sent: Friday, August 28, 2026 8:01 AM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Asim, Thanks for bringing up the DARPA Explainable AI program. I was invited to submit an article to a Special Issue about Explainable AI that was published when the DARPA program was announced. The editors of the Special Issue knew that Adaptive Resonance Theory, or ART, models automatically discover and incrementally learn over multiple learning trials to pay attention to the predictive, or explainable, critical feature patterns that control successful predictions and are used in learning by the adaptive weights in bottom-up adaptive filters and top-down expectations. ART does this AUTOMATICALLY. You seem to have to use an external teacher to do it. Here is the article: Grossberg, S. (2020). A path towards Explainable AI and autonomous adaptive intelligence: Deep Learning, Adaptive Resonance, and models of perception, emotion, and action. Frontiers in Neurobotics, June 25, 2020. https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2020.00036/full The article?s Abstract illustrates its explanatory range [boldface mine]: * "Biological neural network models whereby brains make minds help to understand autonomous adaptive intelligence. This article summarizes why the dynamics and emergent properties of such models for perception, cognition, emotion, and action are explainable, and thus amenable to being confidently implemented in large-scale applications. Key to their explainability is how these models combine fast activations, or short-term memory (STM) traces, and learned weights, or long-term memory (LTM) traces. Visual and auditory perceptual models have explainable conscious STM representations of visual surfaces and auditory streams in surface-shroud resonances and stream-shroud resonances, respectively. Deep Learning is often used to classify data. However, Deep Learning can experience catastrophic forgetting: At any stage of learning, an unpredictable part of its memory can collapse. Even if it makes some accurate classifications, they are not explainable and thus cannot be used with confidence. Deep Learning shares these problems with the back propagation algorithm, whose computational problems due to non-local weight transport during mismatch learning were described in the 1980s. Deep Learning became popular after very fast computers and huge online databases became available that enabled new applications despite these problems. Adaptive Resonance Theory, or ART, algorithms overcome the computational problems of back propagation and Deep Learning. ART is a self-organizing production system that incrementally learns, using arbitrary combinations of unsupervised and supervised learning and only locally computable quantities, to rapidly classify large non-stationary databases without experiencing catastrophic forgetting. ART classifications and predictions are explainable using the attended critical feature patterns in STM on which they build. The LTM adaptive weights of the fuzzy ARTMAP algorithm induce fuzzy IF-THEN rules that explain what feature combinations predict successful outcomes. ART has been successfully used in multiple large-scale real-world applications, including remote sensing, medical database prediction, and social media data clustering. Also explainable are the MOTIVATOR model of reinforcement learning and cognitive-emotional interactions, and the VITE, DIRECT, DIVA, and SOVEREIGN models for reaching, speech production, spatial navigation, and autonomous adaptive intelligence. These biological models exemplify complementary computing, and use local laws for match learning and mismatch learning that avoid the problems of Deep Learning." Chapter 5 in my 2021 Magnum Opus CONSCIOUS MIND, RESONANT BRAIN: HOW EACH BRAIN MAKES A MIND https://www.amazon.com/dp/0190070552?lv=shuf&channelId=500&plpRedirect=mhFallback provides a self-contained and non-technical overview and synthesis of my results about ART and Explainable AI. The other chapters summarize neural network models of the main processes whereby our brains make our conscious minds in healthy individuals and clinical patients. Best, Steve Stephen Grossberg Wang Professor of Cognitive and Neural Systems Director, Center for Adaptive Systems Emeritus Professor of Mathematics & Statistics, Psychological & Brain Sciences, and Biomedical Engineering Boston University sites.bu.edu/steveg/ steve at bu.edu http://en.wikipedia.org/wiki/Stephen_Grossberg http://scholar.google.com/citations?user=3BIV70wAAAAJ&hl=en https://sites.bu.edu/steveg/files/2021/08/Grossberg-CV-8-14-21.pdf https://youtu.be/9n5AnvFur7I https://www.youtube.com/watch?v=_hBye6JQCh4 https://www.amazon.com/Conscious-Mind-Resonant-Brain-Makes/dp/0190070552 https://www.amazon.com/Your-Creative-Brain-Consciously-Experience/dp/0198965370 From: Connectionists > on behalf of Asim Roy > Date: Friday, August 28, 2026 at 3:51?AM To: Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: AI at 50 videos... Stephen, You raise important issues. There are many ways of dealing with them. Here?s one of the ways. The image below shows what DARPA wanted for Explainable AI. Essentially, find the parts of an object for its prediction. So, for a cat, verify some of its unique parts like the fur, whiskers, and claws. Thus, in our explainable models, we ?teach? an object detection model about these parts. What we get as outputs, for a cat, are predictions for the cat and its parts. There?s a layer of logic that uses these outputs and verifies the existence of the parts before ?finally? predicting that there?s a cat in the image. In essence, it?s neuro-symbolic. One can do the symbolic part with a graphical model too. We can also predict that there?s a cat even if we don?t ?see? all of the parts, such as when we just see its face and not the rest of the body. And, in a similar way, your ?morning coffee? concept can be built from its parts that you mention. There?s a long tradition in computer vision of finding parts of objects inside the model. Our initial conception was that the higher-level filters in a CNN correspond to certain abstractions such as a nose, eyes, ears for a human. That was the distributed representation but based on lower-level abstractions (parts of objects). But years of research showed that one could not find these kinds of lower-level abstractions in standard CNN models. So, what do you do? Well, one way is to force some of the filters to correspond to these parts. There?s a long history to this body of work in computer vision, including that of Hinton, to create abstractions within a model. Both ways, we are ?teaching? the models about ?low-level abstraction.? In a way, in both ways, it?s a hierarchical system. In both ways, we create a neuro-symbolic system. By the way, abstractions are generalizations, they don?t code specific episodic memories. Your ?table,? ?coffee,? ?cup? and all that are abstract notions. Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [A screenshot of a computer AI-generated content may be incorrect.] From: Stephen Jos? Hanson > Sent: Thursday, August 27, 2026 4:26 AM To: Asim Roy >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Asim, I sit here with my coffee neuron and my cup neuron creating a cup of coffee, and my table neuron holding the cup and coffee neurons representing a cup of coffee. On the table, on the floor, noting of course the cup of coffee is not on the floor but on the table. So this is a network of coffee, cups, tables, maybe coded by another neuron in a hierarchical network called morning coffee? Well this is the typical problem with localist model of the brain. How would it work exactly? So far just representing a single episodic event requires hierarchical network of neurons that generalize to what exactly? Do we need many such networks to represent all episodic memories of having a cup of coffee? This is similar to the problem with the socalled "fusiform face area" which with some proper controls doesn't really exist (cf Hanson 2022). What would a small blueberry are in fusiform gyrus be doing with all those faces we recognize? A brain face cloud server? Other accounts show that FG, LO, IN and PFC are at least involved in a network of areas (not neurons) in recognizing a face. But computationally what's the actual underlying functions? The shift to "neurons represent symbols' is not helpful to explain meaning. Remember, we have billions of neurons and 100 trillions connections between them. I'm not saying symbols aren't important and Harnad's Grounding arguments are becoming more critical (and need to be made more explicit) as LLMs start to be embedded in robots. Therefore an arguably simpler task, is to identify "symbols" in LLMs since we have no idea how they work either. Are symbols in their "neurons"? Cheers Stephen On 8/24/26 22:56, Asim Roy wrote: 1. Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? 2. Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? 3. Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195?223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -- [cid:image003.jpg at 01DD3895.457E71A0] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image001.png Type: image/png Size: 333780 bytes Desc: image001.png URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image002.jpg Type: image/jpeg Size: 45946 bytes Desc: image002.jpg URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image003.jpg Type: image/jpeg Size: 57383 bytes Desc: image003.jpg URL: From ludovico.montalcini at gmail.com Sun Aug 30 07:56:55 2026 From: ludovico.montalcini at gmail.com (Ludovico Montalcini) Date: Sun, 30 Aug 2026 13:56:55 +0200 Subject: Connectionists: =?utf-8?q?=F0=9F=8C=9F_Last_Seats_Available!_?= =?utf-8?q?=F0=9F=8C=9F_International_Artificial_Intelligence_Summe?= =?utf-8?q?r_School_=E2=80=94_IAISS_2026_5_Days=2E_World-Class_Expe?= =?utf-8?q?rts=2E_Tuscany=2E?= Message-ID: International Artificial Intelligence Summer School ? IAISS 20265 Days. World-Class Experts. Tuscany. *Apologies for multiple copies ? please forward to anyone who might be interested!* ------------------------------ ? A Full-Immersion Experience in Frontier AI Join us for an unforgettable week of cutting-edge lectures on AI and Generative AI, delivered by some of the most renowned researchers in the field ? set against the stunning backdrop of the Tuscan coast. ? *Riva del Sole Resort & SPA*, Castiglione della Pescaia, Tuscany, Italy ? *September 20?24, 2026* ? 2026.iaiss.cc?? iaiss at icas.cc ------------------------------ ? Meet Your Lecturers An extraordinary faculty from leading institutions across the globe ? each delivering up to four lectures on the frontiers of AI research: - *Stefano V. Albrecht* ? Nanyang Technological University, Singapore - *Rafal Bogacz* ? University of Oxford, UK - *Giuseppe De Giacomo* ? University of Oxford, UK - *Christoph Feichtenhofer* ? Meta Superintelligence Labs, USA - *Sven Giesselbach* ? T-Systems International, Germany - *Jamie Hayes* ? Google DeepMind, UK - *Sumi Helal* ? University of Bologna, Italy - *Albert Q. Jiang* ? Mistral AI, Paris & University of Cambridge, UK - *Marine Le Morvan* ? INRIA Saclay, France - *Bruno Lepri* ? FBK, Italy - *Sebastian Risi* ? Sakana AI, Japan & IT University of Copenhagen, Denmark - *George Paliouras* ? NCSR "Demokritos", Athens, Greece - *Panos Pardalos* ? University of Florida, USA ? Explore the full lecturer bios ? Browse the lecture program ------------------------------ ? Present Your Own Research Want to share your work with a world-class audience? Submit a *research abstract* (max 2 pages, any format) during registration. Our scientific committee will select contributions for *poster presentations* and/or *short talks* ? plus a chance to compete for the *IAISS Norbert Wiener Best Presentation Award in Artificial Intelligence*. ? Abstract deadline: *September 6* ? See the Best Presentation Award ------------------------------ ? Why Attend? - *36?40 hours* of intensive lectures from top-tier researchers - Earn *8 ECTS points* (PhD students and highly motivated Master's/BSc students) - Network with peers in a *convivial, professional, and inspiring* setting - A rare opportunity to learn directly from ? and mingle with ? the people shaping the future of AI *Who should attend:* PhD students, PostDocs, junior academics (up to Assistant Professor level), industry practitioners, and exceptionally motivated Master's/BSc students. *Language:* English ------------------------------ ? The Venue *Riva del Sole Resort & SPA* Localit? Riva del Sole, Castiglione della Pescaia (Grosseto), 58043, Tuscany, Italy ? +39 0564 928111 | ?? booking.events at rivadelsole.it ? rivadelsole.it IAISS is a *fully residential* school ? all participants and lecturers stay together at the same resort, creating a truly immersive, community-driven learning experience. No exceptions. *To join us, simply:* 1. *Register* for the school 2. *Book your stay* at Riva del Sole Resort & SPA using the IAISS discounted rates (accommodation form provided at registration) 3. Send your *booking confirmation number* to iaiss at icas.cc *Sharing an apartment?* Send the Hotel your co-guest's name, surname, and email ? otherwise, book a single-use apartment. IAISS is proudly organized as a *non-profit scientific event*. ------------------------------ ? Seats Are Limited ? Register Now! ? *Register here * ? Past editions | ? FAQ | ? Latest news | ? LinkedIn group *See you this September in Riva del Sole!* ? The IAISS 2026 Organizing Committee ? 2026.iaiss.cc | ?? iaiss at icas.cc ------------------------------ *Apologies for multiple copies ? please forward to anyone who might be interested!* -------------- next part -------------- An HTML attachment was scrubbed... URL: From steve at bu.edu Sun Aug 30 18:24:30 2026 From: steve at bu.edu (Grossberg, Stephen) Date: Sun, 30 Aug 2026 22:24:30 +0000 Subject: Connectionists: AI@50 videos...Specialized applications vs. general-purpose self-organizing neural networks for autonomous adaptive intelligence In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> Message-ID: Dear Asim, Your reply is completely consistent with my previous remarks. Keep up the good work! Best, Steve Get Outlook for iOS ________________________________ From: Asim Roy Sent: Sunday, 30 August 2026 18:13:10 To: Grossberg, Stephen ; Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: AI at 50 videos...Specialized applications vs. general-purpose self-organizing neural networks for autonomous adaptive intelligence Dear Steve, As I mentioned earlier, the approach is neurosymbolic. We ?teach? a neuro model about the ?symbols? we are interested in. It?s as simple as that. And we then process the symbolic output in the symbol processing unit. The symbols might be cat and its parts. Or airplanes and ships and their parts. Or drones and their parts. It?s fairly general idea and we can tie it down to how humans are taught to recognize objects. We have not tried the approach in other domains, but, in theory, it should be applicable. We do break down speech and sound into parts. And sentences into words and then syllables. Part-based recognition is powerful. It can handle adversarial attacks quite well. I would consider adversarial attacks as an extension of the concept of invariance. Our approach is model agnostic in the sense that we can as well use your ART family of models as the ?neuro? engine. Right now, we simply use fairly standard off-the-shelf object detection models. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen Sent: Sunday, August 30, 2026 7:37 AM To: Asim Roy ; Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen Subject: Re: Connectionists: AI at 50 videos...Specialized applications vs. general-purpose self-organizing neural networks for autonomous adaptive intelligence Dear Asim, I totally agree that, if you have developed an application that works, use it! Over the years, I and my colleagues have done the same. We have regularly abstracted from our biological neural network models to design fast algorithms for efficient operation in specific application domains. E.g., Srinivasa, N., Bhattacharyya, R., Sundareswara, R., Lee, C., and Grossberg, S. (2012) A bio-inspired kinematic controller for obstacle avoidance during reaching tasks with real robots. Neural Networks, 35, 54-69. https://sites.bu.edu/steveg/files/2016/06/NN_Final.pdf Ivey, R., Bullock, D., and Grossberg, S. (2011). A neuromorphic model of spatial lookahead planning. Neural Networks, 24, 257-266. https://sites.bu.edu/steveg/files/2016/06/IveBulGro2011.pdf Grossberg, S., & Huang, T.-R. (2009). ARTSCENE: A neural system for natural scene classification. Journal of Vision, 9(4):6, 1-19. https://jov.arvojournals.org/article.aspx?articleid=2193487 Hong, S. and Grossberg, S. (2004). A neuromorphic model for achromatic and chromatic surface representation of natural images. Neural Networks, 2004, 17, 787-808. https://sites.bu.edu/steveg/files/2016/06/HongGro2004NN.pdf There are at least two possible problems with such applications: They tend to be domain-specific and, if you try to use them in a different task environment, you often hit a Brick Wall. In my modeling work during the past 69 years of how our brains make our conscious minds, I have never hit a Brick Wall. In fact, earlier principles, mechanisms, and architectures have always formed a foundation for the next discoveries. Indeed, ALL of my models can be expressed in terms of specializations of a few basic equations, and a somewhat larger number of modules or microcircuits, that are assembled into what I call MODAL architectures (for different MODALities), such as vision and image processing; audition, speech, language, and music; attention, categorization, cognitive information processing, and prediction; motivation and cognitive-emotional interactions; adaptive sensory-motor control, including reaching, standing, walking, running, tool use, dancing; and so on. That is why I am still discovering and developing new models, as my web page sites.bu.edu/steveg illustrates. It is also possible, if your current implementations do not achieve all their goals, or the goals change, for you to turn to our models for concepts and mechanisms that you can specialize for your own applications. More generally, taken together, the almost 600 downloadable archival articles on my web page provide a blueprint for the revolutionary computational paradigm of autonomous adaptive intelligence. Self-driving cars are just one example that is currently being developed. In the next century, particularly as societies and technologies need to rapidly adapt to new economic, political, and environmental challenges, increasing autonomous, adaptively intelligent solutions will help us to survive, or at least that is my hope. Best, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 8:32?PM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Steve, I am fully aware that the parts-based explanation has limits. For example, you can?t define the parts of a rock or a boulder easily, if at all, unless you bring in texture, color, and all of that. Part-based explanation is based on human understanding of objects in terms of parts and is consistent with ?teaching? humans about parts of objects. The general idea is to engage with model users and understand which parts they want to verify if a parts-based approach is acceptable for explanation. From our experience, one of the advantages of the part-based approach is that it can in fact deal with occlusions, camouflage, noise, and other kinds of distortions. For example, with this approach, if you just see the face of a cat and the body is occluded (behind a door), you can still recognize the object as a cat and justify with the parts-based logic. And the model can do that even though it was never trained on such images of cats. One can think of the parts-based approach extending the concept of invariance. We can deal with variety. For example, small drones come in different shapes and sizes, and we can do part-based recognition of flying drones. I think we all are aware of your wide body of work, which is extensive. However, I look more closely at the computer vision literature, and they directly address the problem of recognizing parts of objects within their models. Thus, my approach aligns with that body of work. And it?s consistent with the way humans learn and explain. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen > Sent: Saturday, August 29, 2026 4:24 PM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Asim, You write: "make a list of these ?parts? first and then build models based on that list.? This approach was shown decades ago to be unable to deal with novel shapes and their parts. Irv Biederman is perhaps the most distinguished psychologist to have promoted it; e.g., in his 1987 Psychological Review article (94, 115-147) entitled: Recognition-by-Components: A Theory of Human Image Understanding His approach was shown to have only limited explanatory power, since the variety of POSSIBLE novel two-dimensional and three-dimensional parts and wholes exceeds the explanatory range of the kinds of ?models? that you use. Natural objects in the real world cannot easily, if at all, be described by your approach. The speed with which you replied to my extensive list of article titles and abstracts leads me to think that you did not have enough time to even scan them. They offer rigorous computational solutions to the problem that you have posed. Please take the time to read at least the article titles and abstracts. Your approach says nothing about how a two-dimensional picture of a shape is consciously seen as the three-dimensional representation of the shape. The Necker Cube is one of hundreds of examples. Your approach also says nothing about how perceptual grouping binds distributed features into parts or wholes, depending upon Gestalt grouping laws, or how figure-ground separation occurs; notably, how we consciously see the unoccluded parts of a partly occluded object, but often only recognize, without seeing, the occluded parts. And so on... Here are a few additional relevant articles: Cohen, M.A. and Grossberg, S. (1990). Unitized recognition codes for parts and wholes: The unique cue in configural discriminations. In M. Commons, R. Herrnstein, S. Kosslyn, and D. Mumford (Eds.), Computational and clinical approaches to pattern recognition and concept formation. Hillsdale, NJ: Erlbaum. https://sites.bu.edu/steveg/files/2016/06/CohenGro1990UniRecogCodes.pdf Grossberg, S. (2016). Cortical dynamics of figure-ground separation in response to 2D pictures and 3D scenes: How V2 combines border ownership, stereoscopic cues, and Gestalt grouping rules. Frontiers in Psychology. 26 January 2016. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2015.02054/full Grossberg, S., and Swaminathan, G. (2004). A laminar cortical model for 3D perception of slanted and curved surfaces and of 2D images: development, attention and bistability. Vision Research, 44, 1147-1187. https://sites.bu.edu/steveg/files/2016/06/GroSwa2004VR.pdf Congratulations on winning the best article of 2025 award in the magazine of the Military Operations Research Society! Our models have been deployed in software and hardware by MIT Lincoln Lab and other National Labs. We were awarded millions of dollars of competitive grants and contracts by essentially all the major funding agencies in Washington and elsewhere, including ARO, AFOSR, ONR, DARPA, NSF, NIH, Hughes Research Lab, etc. to develop them. We won millions of dollars in grants and contracts because our models work in the real world. Best again, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 3:40?PM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Steve, One can certainly claim that ?views? are ?parts.? It will be hard to dispute that. It?s the labeling of these ?parts? that?s the issue. In our Explanation First approach, we make a list of these ?parts? first and then build models based on that list. We have used this approach, for example, to identify different types of ships and aircraft from satellite images. Some standard parts for aircraft are like the fuselage, wings, tail and so on. The Explanation First approach is very similar to an architect understanding the needs of his/her client before designing a house. The ?Explanation First, Models Next? approach is model agnostic. So, we can use one of the ARTMAP models as the underlying model. There is no conflict there. By the way, I explained this Explanation First approach in a simple article in Phalanx last year, the magazine of the Military Operations Research Society (Home - Military Operations Research Society) and it was awarded the best article of 2025. We are starting to use the approach in certain military applications. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen > Sent: Saturday, August 29, 2026 6:25 AM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Asim, Models of the ARTMAP family (fuzzy ARTMAP, distributed ARTMAP, Gaussian ARTMAP, etc.) are perhaps the simplest kind of self-organizing neural model that can use a teacher (typically the environment, but it can also be a human) to learn to recognize objects, their parts, and their names. The work that my colleagues and I have published has gone far beyond ARTMAP. We have, in a series of articles that can be downloaded from sites.bu.edu/steveg, shown how humans can learn view-, position-, and size-invariant object representations, and use them to carry out actions that achieve valued goals. Our models show how both view-specific and view-invariant representations of an object (say, a particular view of your mother?s face AND an invariant one; namely, that she has a face) can simultaneously guide recognition of mom AND specific actions in response to where the view of her face is looking. The ?views? are ?parts?. I will list some of the relevant article titles and their urls here, for your convenience. Later articles tend to model ever more challenging aspects of self-organizing object recognition and action. DARPA gave us large grants over the years to do this work. Carpenter, G.A., Grossberg, S., and Mehanian, C. (1989). Invariant recognition of cluttered scenes by a self- organizing ART architecture: CORT-X boundary segmentation. Neural Networks, 2, 169-181. https://sites.bu.edu/steveg/files/2016/06/CarGroMeh1989NN.pdf Grossberg, S. and Wyse, L. (1991). Invariant recognition of cluttered scenes by a self-organizing ART architecture: Figure-ground separation. Neural Networks, 1991, 4, 723-742. https://sites.bu.edu/steveg/files/2016/06/GroWyse1991NN.pdf Fazl, A., Grossberg, S., and Mingolla, E. (2009). View-invariant object category learning, recognition, and search: How spatial and object attention are coordinated using surface-based attentional shrouds. Cognitive Psychology, 58, 1-48. https://sites.bu.edu/steveg/files/2016/06/FazGroMin2008.pdf Grossberg, S., and Vladusich, T. (2010). How do children learn to follow gaze, share joint attention, imitate their teachers, and use tools during social interactions? Neural Networks, 23, 940-965. https://sites.bu.edu/steveg/files/2016/06/GrossbergVladusichNN2010.pdf Cao, Y., Grossberg, S., and Markowitz, J. (2011). How does the brain rapidly learn and reorganize view- and positionally-invariant object representations in inferior temporal cortex? Neural Networks, 24, 1050-1061. https://sites.bu.edu/steveg/files/2016/06/NN2853.pdf Grossberg, S., Markowitz, J., and Cao, Y. (2011). On the road to invariant recognition: Explaining tradeoff and morph properties of cells in inferotemporal cortex using multiple-scale task-sensitive attentive learning. Neural Networks, 24, 1036-1049. https://sites.bu.edu/steveg/files/2016/06/GroMarCao2011TR.pdf Grossberg, S., Srinivasan, K., and Yazdabakhsh, A. (2011). On the road to invariant object recognition: How cortical area V2 transforms absolute to relative disparity during 3D vision. Neural Networks, 24, 686-692. https://sites.bu.edu/steveg/files/2016/06/GroSriYaz2011TR.pdf Chang, H.-C., Grossberg, S., and Cao, Y. (2014) Where?s Waldo? How perceptual cognitive, and emotional brain processes cooperate during learning to categorize and find desired objects in a cluttered scene. Frontiers in Integrative Neuroscience, doi: 10.3389/fnint.2014.0043, https://www.frontiersin.org/journals/integrative-neuroscience/articles/10.3389/fnint.2014.00043/full Grossberg, S., Srinivasan, K., and Yazdanbakhsh, A. (2014). Binocular fusion and invariant category learning due to predictive remapping during scanning of a depthful scene with eye movements. Frontiers in Psychology: Perception Science, doi: 10.3389/fpsyg.2014.01457. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.01457/full Grossberg, S. (2019). The embodied brain of SOVEREIGN2: From space-variant conscious percepts during visual search and navigation to learning invariant object categories and cognitive-emotional plans for acquiring valued goals. Frontiers in Computational Neuroscience. Published online: June 25, 2019. https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2019.00036/full Best, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 5:09?AM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Steve, What DARPA seemed to want at that time for computer vision/object recognition is object prediction based on part verification ? if a cat, have you verified its unique parts? That?s the kind of explanation they wanted. ?Teaching? an object recognition system about these parts (to generalize, parts are just subconcepts) is not inconsistent with the way we teach humans. For example, a one- or two-year-old recognizes a car. But we teach them about parts much later - such as what a tire is, a door, steering wheel and so on. How a person?s internal biological model is adjusted from these ?part teachings? I am not sure, but the identification and teaching (labeling) about parts is very much a part of human learning. One might say that LLMs automate this type of learning, but we do provide labels in the descriptions of objects. So, that?s ?teaching.? When we started with multilayered models, the assumption was that at higher levels, there would be part abstractions such as those shown in the figure below - the dog face, legs, tail and so on. The problem was, we couldn?t find these abstractions at higher levels of the network. Even if you found them, we would need to assign labels or names to them ? a dog face, legs, tail and so on. Once you get to assigning these labels, that?s ?teaching.? There is a body of work in computer vision trying to build these abstractions inside CNN models. In our method, we also ?teach? our deep learning models, but in a different way. In general, ?teaching? would be necessary whichever way you do it. We call our approach ?Explanation First, Model Next.? We have reversed the process of model building. Our testing shows gain in overall accuracy compared to some standard object detection models. This DARPA style approach also enables resistance against adversarial attacks. So substantial benefits with the Explanation First approach. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [cid:image001.png at 01DD388F.878E4670] From: Grossberg, Stephen > Sent: Friday, August 28, 2026 8:01 AM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Asim, Thanks for bringing up the DARPA Explainable AI program. I was invited to submit an article to a Special Issue about Explainable AI that was published when the DARPA program was announced. The editors of the Special Issue knew that Adaptive Resonance Theory, or ART, models automatically discover and incrementally learn over multiple learning trials to pay attention to the predictive, or explainable, critical feature patterns that control successful predictions and are used in learning by the adaptive weights in bottom-up adaptive filters and top-down expectations. ART does this AUTOMATICALLY. You seem to have to use an external teacher to do it. Here is the article: Grossberg, S. (2020). A path towards Explainable AI and autonomous adaptive intelligence: Deep Learning, Adaptive Resonance, and models of perception, emotion, and action. Frontiers in Neurobotics, June 25, 2020. https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2020.00036/full The article?s Abstract illustrates its explanatory range [boldface mine]: * "Biological neural network models whereby brains make minds help to understand autonomous adaptive intelligence. This article summarizes why the dynamics and emergent properties of such models for perception, cognition, emotion, and action are explainable, and thus amenable to being confidently implemented in large-scale applications. Key to their explainability is how these models combine fast activations, or short-term memory (STM) traces, and learned weights, or long-term memory (LTM) traces. Visual and auditory perceptual models have explainable conscious STM representations of visual surfaces and auditory streams in surface-shroud resonances and stream-shroud resonances, respectively. Deep Learning is often used to classify data. However, Deep Learning can experience catastrophic forgetting: At any stage of learning, an unpredictable part of its memory can collapse. Even if it makes some accurate classifications, they are not explainable and thus cannot be used with confidence. Deep Learning shares these problems with the back propagation algorithm, whose computational problems due to non-local weight transport during mismatch learning were described in the 1980s. Deep Learning became popular after very fast computers and huge online databases became available that enabled new applications despite these problems. Adaptive Resonance Theory, or ART, algorithms overcome the computational problems of back propagation and Deep Learning. ART is a self-organizing production system that incrementally learns, using arbitrary combinations of unsupervised and supervised learning and only locally computable quantities, to rapidly classify large non-stationary databases without experiencing catastrophic forgetting. ART classifications and predictions are explainable using the attended critical feature patterns in STM on which they build. The LTM adaptive weights of the fuzzy ARTMAP algorithm induce fuzzy IF-THEN rules that explain what feature combinations predict successful outcomes. ART has been successfully used in multiple large-scale real-world applications, including remote sensing, medical database prediction, and social media data clustering. Also explainable are the MOTIVATOR model of reinforcement learning and cognitive-emotional interactions, and the VITE, DIRECT, DIVA, and SOVEREIGN models for reaching, speech production, spatial navigation, and autonomous adaptive intelligence. These biological models exemplify complementary computing, and use local laws for match learning and mismatch learning that avoid the problems of Deep Learning." Chapter 5 in my 2021 Magnum Opus CONSCIOUS MIND, RESONANT BRAIN: HOW EACH BRAIN MAKES A MIND https://www.amazon.com/dp/0190070552?lv=shuf&channelId=500&plpRedirect=mhFallback provides a self-contained and non-technical overview and synthesis of my results about ART and Explainable AI. The other chapters summarize neural network models of the main processes whereby our brains make our conscious minds in healthy individuals and clinical patients. Best, Steve Stephen Grossberg Wang Professor of Cognitive and Neural Systems Director, Center for Adaptive Systems Emeritus Professor of Mathematics & Statistics, Psychological & Brain Sciences, and Biomedical Engineering Boston University sites.bu.edu/steveg/ steve at bu.edu http://en.wikipedia.org/wiki/Stephen_Grossberg http://scholar.google.com/citations?user=3BIV70wAAAAJ&hl=en https://sites.bu.edu/steveg/files/2021/08/Grossberg-CV-8-14-21.pdf https://youtu.be/9n5AnvFur7I https://www.youtube.com/watch?v=_hBye6JQCh4 https://www.amazon.com/Conscious-Mind-Resonant-Brain-Makes/dp/0190070552 https://www.amazon.com/Your-Creative-Brain-Consciously-Experience/dp/0198965370 From: Connectionists > on behalf of Asim Roy > Date: Friday, August 28, 2026 at 3:51?AM To: Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: AI at 50 videos... Stephen, You raise important issues. There are many ways of dealing with them. Here?s one of the ways. The image below shows what DARPA wanted for Explainable AI. Essentially, find the parts of an object for its prediction. So, for a cat, verify some of its unique parts like the fur, whiskers, and claws. Thus, in our explainable models, we ?teach? an object detection model about these parts. What we get as outputs, for a cat, are predictions for the cat and its parts. There?s a layer of logic that uses these outputs and verifies the existence of the parts before ?finally? predicting that there?s a cat in the image. In essence, it?s neuro-symbolic. One can do the symbolic part with a graphical model too. We can also predict that there?s a cat even if we don?t ?see? all of the parts, such as when we just see its face and not the rest of the body. And, in a similar way, your ?morning coffee? concept can be built from its parts that you mention. There?s a long tradition in computer vision of finding parts of objects inside the model. Our initial conception was that the higher-level filters in a CNN correspond to certain abstractions such as a nose, eyes, ears for a human. That was the distributed representation but based on lower-level abstractions (parts of objects). But years of research showed that one could not find these kinds of lower-level abstractions in standard CNN models. So, what do you do? Well, one way is to force some of the filters to correspond to these parts. There?s a long history to this body of work in computer vision, including that of Hinton, to create abstractions within a model. Both ways, we are ?teaching? the models about ?low-level abstraction.? In a way, in both ways, it?s a hierarchical system. In both ways, we create a neuro-symbolic system. By the way, abstractions are generalizations, they don?t code specific episodic memories. Your ?table,? ?coffee,? ?cup? and all that are abstract notions. Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [A screenshot of a computer AI-generated content may be incorrect.] From: Stephen Jos? Hanson > Sent: Thursday, August 27, 2026 4:26 AM To: Asim Roy >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Asim, I sit here with my coffee neuron and my cup neuron creating a cup of coffee, and my table neuron holding the cup and coffee neurons representing a cup of coffee. On the table, on the floor, noting of course the cup of coffee is not on the floor but on the table. So this is a network of coffee, cups, tables, maybe coded by another neuron in a hierarchical network called morning coffee? Well this is the typical problem with localist model of the brain. How would it work exactly? So far just representing a single episodic event requires hierarchical network of neurons that generalize to what exactly? Do we need many such networks to represent all episodic memories of having a cup of coffee? This is similar to the problem with the socalled "fusiform face area" which with some proper controls doesn't really exist (cf Hanson 2022). What would a small blueberry are in fusiform gyrus be doing with all those faces we recognize? A brain face cloud server? Other accounts show that FG, LO, IN and PFC are at least involved in a network of areas (not neurons) in recognizing a face. But computationally what's the actual underlying functions? The shift to "neurons represent symbols' is not helpful to explain meaning. Remember, we have billions of neurons and 100 trillions connections between them. I'm not saying symbols aren't important and Harnad's Grounding arguments are becoming more critical (and need to be made more explicit) as LLMs start to be embedded in robots. Therefore an arguably simpler task, is to identify "symbols" in LLMs since we have no idea how they work either. Are symbols in their "neurons"? Cheers Stephen On 8/24/26 22:56, Asim Roy wrote: 1. Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? 2. Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? 3. Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195?223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -- [cid:image003.jpg at 01DD388F.878E4670] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image001.png Type: image/png Size: 333780 bytes Desc: image001.png URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image002.jpg Type: image/jpeg Size: 45946 bytes Desc: image002.jpg URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image003.jpg Type: image/jpeg Size: 57383 bytes Desc: image003.jpg URL: From steve at bu.edu Sat Aug 29 19:24:15 2026 From: steve at bu.edu (Grossberg, Stephen) Date: Sat, 29 Aug 2026 23:24:15 +0000 Subject: Connectionists: AI@50 videos...Self-organized learning of objects and their parts In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> Message-ID: Dear Asim, You write: "make a list of these ?parts? first and then build models based on that list.? This approach was shown decades ago to be unable to deal with novel shapes and their parts. Irv Biederman is perhaps the most distinguished psychologist to have promoted it; e.g., in his 1987 Psychological Review article (94, 115-147) entitled: Recognition-by-Components: A Theory of Human Image Understanding His approach was shown to have only limited explanatory power, since the variety of POSSIBLE novel two-dimensional and three-dimensional parts and wholes exceeds the explanatory range of the kinds of ?models? that you use. Natural objects in the real world cannot easily, if at all, be described by your approach. The speed with which you replied to my extensive list of article titles and abstracts leads me to think that you did not have enough time to even scan them. They offer rigorous computational solutions to the problem that you have posed. Please take the time to read at least the article titles and abstracts. Your approach says nothing about how a two-dimensional picture of a shape is consciously seen as the three-dimensional representation of the shape. The Necker Cube is one of hundreds of examples. Your approach also says nothing about how perceptual grouping binds distributed features into parts or wholes, depending upon Gestalt grouping laws, or how figure-ground separation occurs; notably, how we consciously see the unoccluded parts of a partly occluded object, but often only recognize, without seeing, the occluded parts. And so on... Here are a few additional relevant articles: Cohen, M.A. and Grossberg, S. (1990). Unitized recognition codes for parts and wholes: The unique cue in configural discriminations. In M. Commons, R. Herrnstein, S. Kosslyn, and D. Mumford (Eds.), Computational and clinical approaches to pattern recognition and concept formation. Hillsdale, NJ: Erlbaum. https://sites.bu.edu/steveg/files/2016/06/CohenGro1990UniRecogCodes.pdf Grossberg, S. (2016). Cortical dynamics of figure-ground separation in response to 2D pictures and 3D scenes: How V2 combines border ownership, stereoscopic cues, and Gestalt grouping rules. Frontiers in Psychology. 26 January 2016. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2015.02054/full Grossberg, S., and Swaminathan, G. (2004). A laminar cortical model for 3D perception of slanted and curved surfaces and of 2D images: development, attention and bistability. Vision Research, 44, 1147-1187. https://sites.bu.edu/steveg/files/2016/06/GroSwa2004VR.pdf Congratulations on winning the best article of 2025 award in the magazine of the Military Operations Research Society! Our models have been deployed in software and hardware by MIT Lincoln Lab and other National Labs. We were awarded millions of dollars of competitive grants and contracts by essentially all the major funding agencies in Washington and elsewhere, including ARO, AFOSR, ONR, DARPA, NSF, NIH, Hughes Research Lab, etc. to develop them. We won millions of dollars in grants and contracts because our models work in the real world. Best again, Steve Get Outlook for Mac From: Asim Roy Date: Saturday, August 29, 2026 at 3:40?PM To: Grossberg, Stephen ; Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Steve, One can certainly claim that ?views? are ?parts.? It will be hard to dispute that. It?s the labeling of these ?parts? that?s the issue. In our Explanation First approach, we make a list of these ?parts? first and then build models based on that list. We have used this approach, for example, to identify different types of ships and aircraft from satellite images. Some standard parts for aircraft are like the fuselage, wings, tail and so on. The Explanation First approach is very similar to an architect understanding the needs of his/her client before designing a house. The ?Explanation First, Models Next? approach is model agnostic. So, we can use one of the ARTMAP models as the underlying model. There is no conflict there. By the way, I explained this Explanation First approach in a simple article in Phalanx last year, the magazine of the Military Operations Research Society (Home - Military Operations Research Society) and it was awarded the best article of 2025. We are starting to use the approach in certain military applications. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen Sent: Saturday, August 29, 2026 6:25 AM To: Asim Roy ; Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen Subject: Re: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Asim, Models of the ARTMAP family (fuzzy ARTMAP, distributed ARTMAP, Gaussian ARTMAP, etc.) are perhaps the simplest kind of self-organizing neural model that can use a teacher (typically the environment, but it can also be a human) to learn to recognize objects, their parts, and their names. The work that my colleagues and I have published has gone far beyond ARTMAP. We have, in a series of articles that can be downloaded from sites.bu.edu/steveg, shown how humans can learn view-, position-, and size-invariant object representations, and use them to carry out actions that achieve valued goals. Our models show how both view-specific and view-invariant representations of an object (say, a particular view of your mother?s face AND an invariant one; namely, that she has a face) can simultaneously guide recognition of mom AND specific actions in response to where the view of her face is looking. The ?views? are ?parts?. I will list some of the relevant article titles and their urls here, for your convenience. Later articles tend to model ever more challenging aspects of self-organizing object recognition and action. DARPA gave us large grants over the years to do this work. Carpenter, G.A., Grossberg, S., and Mehanian, C. (1989). Invariant recognition of cluttered scenes by a self- organizing ART architecture: CORT-X boundary segmentation. Neural Networks, 2, 169-181. https://sites.bu.edu/steveg/files/2016/06/CarGroMeh1989NN.pdf Grossberg, S. and Wyse, L. (1991). Invariant recognition of cluttered scenes by a self-organizing ART architecture: Figure-ground separation. Neural Networks, 1991, 4, 723-742. https://sites.bu.edu/steveg/files/2016/06/GroWyse1991NN.pdf Fazl, A., Grossberg, S., and Mingolla, E. (2009). View-invariant object category learning, recognition, and search: How spatial and object attention are coordinated using surface-based attentional shrouds. Cognitive Psychology, 58, 1-48. https://sites.bu.edu/steveg/files/2016/06/FazGroMin2008.pdf Grossberg, S., and Vladusich, T. (2010). How do children learn to follow gaze, share joint attention, imitate their teachers, and use tools during social interactions? Neural Networks, 23, 940-965. https://sites.bu.edu/steveg/files/2016/06/GrossbergVladusichNN2010.pdf Cao, Y., Grossberg, S., and Markowitz, J. (2011). How does the brain rapidly learn and reorganize view- and positionally-invariant object representations in inferior temporal cortex? Neural Networks, 24, 1050-1061. https://sites.bu.edu/steveg/files/2016/06/NN2853.pdf Grossberg, S., Markowitz, J., and Cao, Y. (2011). On the road to invariant recognition: Explaining tradeoff and morph properties of cells in inferotemporal cortex using multiple-scale task-sensitive attentive learning. Neural Networks, 24, 1036-1049. https://sites.bu.edu/steveg/files/2016/06/GroMarCao2011TR.pdf Grossberg, S., Srinivasan, K., and Yazdabakhsh, A. (2011). On the road to invariant object recognition: How cortical area V2 transforms absolute to relative disparity during 3D vision. Neural Networks, 24, 686-692. https://sites.bu.edu/steveg/files/2016/06/GroSriYaz2011TR.pdf Chang, H.-C., Grossberg, S., and Cao, Y. (2014) Where?s Waldo? How perceptual cognitive, and emotional brain processes cooperate during learning to categorize and find desired objects in a cluttered scene. Frontiers in Integrative Neuroscience, doi: 10.3389/fnint.2014.0043, https://www.frontiersin.org/journals/integrative-neuroscience/articles/10.3389/fnint.2014.00043/full Grossberg, S., Srinivasan, K., and Yazdanbakhsh, A. (2014). Binocular fusion and invariant category learning due to predictive remapping during scanning of a depthful scene with eye movements. Frontiers in Psychology: Perception Science, doi: 10.3389/fpsyg.2014.01457. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.01457/full Grossberg, S. (2019). The embodied brain of SOVEREIGN2: From space-variant conscious percepts during visual search and navigation to learning invariant object categories and cognitive-emotional plans for acquiring valued goals. Frontiers in Computational Neuroscience. Published online: June 25, 2019. https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2019.00036/full Best, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 5:09?AM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Steve, What DARPA seemed to want at that time for computer vision/object recognition is object prediction based on part verification ? if a cat, have you verified its unique parts? That?s the kind of explanation they wanted. ?Teaching? an object recognition system about these parts (to generalize, parts are just subconcepts) is not inconsistent with the way we teach humans. For example, a one- or two-year-old recognizes a car. But we teach them about parts much later - such as what a tire is, a door, steering wheel and so on. How a person?s internal biological model is adjusted from these ?part teachings? I am not sure, but the identification and teaching (labeling) about parts is very much a part of human learning. One might say that LLMs automate this type of learning, but we do provide labels in the descriptions of objects. So, that?s ?teaching.? When we started with multilayered models, the assumption was that at higher levels, there would be part abstractions such as those shown in the figure below - the dog face, legs, tail and so on. The problem was, we couldn?t find these abstractions at higher levels of the network. Even if you found them, we would need to assign labels or names to them ? a dog face, legs, tail and so on. Once you get to assigning these labels, that?s ?teaching.? There is a body of work in computer vision trying to build these abstractions inside CNN models. In our method, we also ?teach? our deep learning models, but in a different way. In general, ?teaching? would be necessary whichever way you do it. We call our approach ?Explanation First, Model Next.? We have reversed the process of model building. Our testing shows gain in overall accuracy compared to some standard object detection models. This DARPA style approach also enables resistance against adversarial attacks. So substantial benefits with the Explanation First approach. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [cid:image003.png at 01DD37B1.AF6B0B30] From: Grossberg, Stephen > Sent: Friday, August 28, 2026 8:01 AM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Asim, Thanks for bringing up the DARPA Explainable AI program. I was invited to submit an article to a Special Issue about Explainable AI that was published when the DARPA program was announced. The editors of the Special Issue knew that Adaptive Resonance Theory, or ART, models automatically discover and incrementally learn over multiple learning trials to pay attention to the predictive, or explainable, critical feature patterns that control successful predictions and are used in learning by the adaptive weights in bottom-up adaptive filters and top-down expectations. ART does this AUTOMATICALLY. You seem to have to use an external teacher to do it. Here is the article: Grossberg, S. (2020). A path towards Explainable AI and autonomous adaptive intelligence: Deep Learning, Adaptive Resonance, and models of perception, emotion, and action. Frontiers in Neurobotics, June 25, 2020. https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2020.00036/full The article?s Abstract illustrates its explanatory range [boldface mine]: * "Biological neural network models whereby brains make minds help to understand autonomous adaptive intelligence. This article summarizes why the dynamics and emergent properties of such models for perception, cognition, emotion, and action are explainable, and thus amenable to being confidently implemented in large-scale applications. Key to their explainability is how these models combine fast activations, or short-term memory (STM) traces, and learned weights, or long-term memory (LTM) traces. Visual and auditory perceptual models have explainable conscious STM representations of visual surfaces and auditory streams in surface-shroud resonances and stream-shroud resonances, respectively. Deep Learning is often used to classify data. However, Deep Learning can experience catastrophic forgetting: At any stage of learning, an unpredictable part of its memory can collapse. Even if it makes some accurate classifications, they are not explainable and thus cannot be used with confidence. Deep Learning shares these problems with the back propagation algorithm, whose computational problems due to non-local weight transport during mismatch learning were described in the 1980s. Deep Learning became popular after very fast computers and huge online databases became available that enabled new applications despite these problems. Adaptive Resonance Theory, or ART, algorithms overcome the computational problems of back propagation and Deep Learning. ART is a self-organizing production system that incrementally learns, using arbitrary combinations of unsupervised and supervised learning and only locally computable quantities, to rapidly classify large non-stationary databases without experiencing catastrophic forgetting. ART classifications and predictions are explainable using the attended critical feature patterns in STM on which they build. The LTM adaptive weights of the fuzzy ARTMAP algorithm induce fuzzy IF-THEN rules that explain what feature combinations predict successful outcomes. ART has been successfully used in multiple large-scale real-world applications, including remote sensing, medical database prediction, and social media data clustering. Also explainable are the MOTIVATOR model of reinforcement learning and cognitive-emotional interactions, and the VITE, DIRECT, DIVA, and SOVEREIGN models for reaching, speech production, spatial navigation, and autonomous adaptive intelligence. These biological models exemplify complementary computing, and use local laws for match learning and mismatch learning that avoid the problems of Deep Learning." Chapter 5 in my 2021 Magnum Opus CONSCIOUS MIND, RESONANT BRAIN: HOW EACH BRAIN MAKES A MIND https://www.amazon.com/dp/0190070552?lv=shuf&channelId=500&plpRedirect=mhFallback provides a self-contained and non-technical overview and synthesis of my results about ART and Explainable AI. The other chapters summarize neural network models of the main processes whereby our brains make our conscious minds in healthy individuals and clinical patients. Best, Steve Stephen Grossberg Wang Professor of Cognitive and Neural Systems Director, Center for Adaptive Systems Emeritus Professor of Mathematics & Statistics, Psychological & Brain Sciences, and Biomedical Engineering Boston University sites.bu.edu/steveg/ steve at bu.edu http://en.wikipedia.org/wiki/Stephen_Grossberg http://scholar.google.com/citations?user=3BIV70wAAAAJ&hl=en https://sites.bu.edu/steveg/files/2021/08/Grossberg-CV-8-14-21.pdf https://youtu.be/9n5AnvFur7I https://www.youtube.com/watch?v=_hBye6JQCh4 https://www.amazon.com/Conscious-Mind-Resonant-Brain-Makes/dp/0190070552 https://www.amazon.com/Your-Creative-Brain-Consciously-Experience/dp/0198965370 From: Connectionists > on behalf of Asim Roy > Date: Friday, August 28, 2026 at 3:51?AM To: Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: AI at 50 videos... Stephen, You raise important issues. There are many ways of dealing with them. Here?s one of the ways. The image below shows what DARPA wanted for Explainable AI. Essentially, find the parts of an object for its prediction. So, for a cat, verify some of its unique parts like the fur, whiskers, and claws. Thus, in our explainable models, we ?teach? an object detection model about these parts. What we get as outputs, for a cat, are predictions for the cat and its parts. There?s a layer of logic that uses these outputs and verifies the existence of the parts before ?finally? predicting that there?s a cat in the image. In essence, it?s neuro-symbolic. One can do the symbolic part with a graphical model too. We can also predict that there?s a cat even if we don?t ?see? all of the parts, such as when we just see its face and not the rest of the body. And, in a similar way, your ?morning coffee? concept can be built from its parts that you mention. There?s a long tradition in computer vision of finding parts of objects inside the model. Our initial conception was that the higher-level filters in a CNN correspond to certain abstractions such as a nose, eyes, ears for a human. That was the distributed representation but based on lower-level abstractions (parts of objects). But years of research showed that one could not find these kinds of lower-level abstractions in standard CNN models. So, what do you do? Well, one way is to force some of the filters to correspond to these parts. There?s a long history to this body of work in computer vision, including that of Hinton, to create abstractions within a model. Both ways, we are ?teaching? the models about ?low-level abstraction.? In a way, in both ways, it?s a hierarchical system. In both ways, we create a neuro-symbolic system. By the way, abstractions are generalizations, they don?t code specific episodic memories. Your ?table,? ?coffee,? ?cup? and all that are abstract notions. Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [A screenshot of a computer AI-generated content may be incorrect.] From: Stephen Jos? Hanson > Sent: Thursday, August 27, 2026 4:26 AM To: Asim Roy >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Asim, I sit here with my coffee neuron and my cup neuron creating a cup of coffee, and my table neuron holding the cup and coffee neurons representing a cup of coffee. On the table, on the floor, noting of course the cup of coffee is not on the floor but on the table. So this is a network of coffee, cups, tables, maybe coded by another neuron in a hierarchical network called morning coffee? Well this is the typical problem with localist model of the brain. How would it work exactly? So far just representing a single episodic event requires hierarchical network of neurons that generalize to what exactly? Do we need many such networks to represent all episodic memories of having a cup of coffee? This is similar to the problem with the socalled "fusiform face area" which with some proper controls doesn't really exist (cf Hanson 2022). What would a small blueberry are in fusiform gyrus be doing with all those faces we recognize? A brain face cloud server? Other accounts show that FG, LO, IN and PFC are at least involved in a network of areas (not neurons) in recognizing a face. But computationally what's the actual underlying functions? The shift to "neurons represent symbols' is not helpful to explain meaning. Remember, we have billions of neurons and 100 trillions connections between them. I'm not saying symbols aren't important and Harnad's Grounding arguments are becoming more critical (and need to be made more explicit) as LLMs start to be embedded in robots. Therefore an arguably simpler task, is to identify "symbols" in LLMs since we have no idea how they work either. Are symbols in their "neurons"? Cheers Stephen On 8/24/26 22:56, Asim Roy wrote: 1. Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? 2. Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? 3. Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195?223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -- [cid:image005.jpg at 01DD37B1.AF6B0B30] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image003.png Type: image/png Size: 333780 bytes Desc: image003.png URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image004.jpg Type: image/jpeg Size: 45946 bytes Desc: image004.jpg URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image005.jpg Type: image/jpeg Size: 57383 bytes Desc: image005.jpg URL: From steve at bu.edu Sun Aug 30 10:36:40 2026 From: steve at bu.edu (Grossberg, Stephen) Date: Sun, 30 Aug 2026 14:36:40 +0000 Subject: Connectionists: AI@50 videos...Specialized applications vs. general-purpose self-organizing neural networks for autonomous adaptive intelligence In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> Message-ID: Dear Asim, I totally agree that, if you have developed an application that works, use it! Over the years, I and my colleagues have done the same. We have regularly abstracted from our biological neural network models to design fast algorithms for efficient operation in specific application domains. E.g., Srinivasa, N., Bhattacharyya, R., Sundareswara, R., Lee, C., and Grossberg, S. (2012) A bio-inspired kinematic controller for obstacle avoidance during reaching tasks with real robots. Neural Networks, 35, 54-69. https://sites.bu.edu/steveg/files/2016/06/NN_Final.pdf Ivey, R., Bullock, D., and Grossberg, S. (2011). A neuromorphic model of spatial lookahead planning. Neural Networks, 24, 257-266. https://sites.bu.edu/steveg/files/2016/06/IveBulGro2011.pdf Grossberg, S., & Huang, T.-R. (2009). ARTSCENE: A neural system for natural scene classification. Journal of Vision, 9(4):6, 1-19. https://jov.arvojournals.org/article.aspx?articleid=2193487 Hong, S. and Grossberg, S. (2004). A neuromorphic model for achromatic and chromatic surface representation of natural images. Neural Networks, 2004, 17, 787-808. https://sites.bu.edu/steveg/files/2016/06/HongGro2004NN.pdf There are at least two possible problems with such applications: They tend to be domain-specific and, if you try to use them in a different task environment, you often hit a Brick Wall. In my modeling work during the past 69 years of how our brains make our conscious minds, I have never hit a Brick Wall. In fact, earlier principles, mechanisms, and architectures have always formed a foundation for the next discoveries. Indeed, ALL of my models can be expressed in terms of specializations of a few basic equations, and a somewhat larger number of modules or microcircuits, that are assembled into what I call MODAL architectures (for different MODALities), such as vision and image processing; audition, speech, language, and music; attention, categorization, cognitive information processing, and prediction; motivation and cognitive-emotional interactions; adaptive sensory-motor control, including reaching, standing, walking, running, tool use, dancing; and so on. That is why I am still discovering and developing new models, as my web page sites.bu.edu/steveg illustrates. It is also possible, if your current implementations do not achieve all their goals, or the goals change, for you to turn to our models for concepts and mechanisms that you can specialize for your own applications. More generally, taken together, the almost 600 downloadable archival articles on my web page provide a blueprint for the revolutionary computational paradigm of autonomous adaptive intelligence. Self-driving cars are just one example that is currently being developed. In the next century, particularly as societies and technologies need to rapidly adapt to new economic, political, and environmental challenges, increasing autonomous, adaptively intelligent solutions will help us to survive, or at least that is my hope. Best, Steve Get Outlook for Mac From: Asim Roy Date: Saturday, August 29, 2026 at 8:32?PM To: Grossberg, Stephen ; Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu Subject: RE: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Steve, I am fully aware that the parts-based explanation has limits. For example, you can?t define the parts of a rock or a boulder easily, if at all, unless you bring in texture, color, and all of that. Part-based explanation is based on human understanding of objects in terms of parts and is consistent with ?teaching? humans about parts of objects. The general idea is to engage with model users and understand which parts they want to verify if a parts-based approach is acceptable for explanation. From our experience, one of the advantages of the part-based approach is that it can in fact deal with occlusions, camouflage, noise, and other kinds of distortions. For example, with this approach, if you just see the face of a cat and the body is occluded (behind a door), you can still recognize the object as a cat and justify with the parts-based logic. And the model can do that even though it was never trained on such images of cats. One can think of the parts-based approach extending the concept of invariance. We can deal with variety. For example, small drones come in different shapes and sizes, and we can do part-based recognition of flying drones. I think we all are aware of your wide body of work, which is extensive. However, I look more closely at the computer vision literature, and they directly address the problem of recognizing parts of objects within their models. Thus, my approach aligns with that body of work. And it?s consistent with the way humans learn and explain. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen Sent: Saturday, August 29, 2026 4:24 PM To: Asim Roy ; Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen Subject: Re: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Asim, You write: "make a list of these ?parts? first and then build models based on that list.? This approach was shown decades ago to be unable to deal with novel shapes and their parts. Irv Biederman is perhaps the most distinguished psychologist to have promoted it; e.g., in his 1987 Psychological Review article (94, 115-147) entitled: Recognition-by-Components: A Theory of Human Image Understanding His approach was shown to have only limited explanatory power, since the variety of POSSIBLE novel two-dimensional and three-dimensional parts and wholes exceeds the explanatory range of the kinds of ?models? that you use. Natural objects in the real world cannot easily, if at all, be described by your approach. The speed with which you replied to my extensive list of article titles and abstracts leads me to think that you did not have enough time to even scan them. They offer rigorous computational solutions to the problem that you have posed. Please take the time to read at least the article titles and abstracts. Your approach says nothing about how a two-dimensional picture of a shape is consciously seen as the three-dimensional representation of the shape. The Necker Cube is one of hundreds of examples. Your approach also says nothing about how perceptual grouping binds distributed features into parts or wholes, depending upon Gestalt grouping laws, or how figure-ground separation occurs; notably, how we consciously see the unoccluded parts of a partly occluded object, but often only recognize, without seeing, the occluded parts. And so on... Here are a few additional relevant articles: Cohen, M.A. and Grossberg, S. (1990). Unitized recognition codes for parts and wholes: The unique cue in configural discriminations. In M. Commons, R. Herrnstein, S. Kosslyn, and D. Mumford (Eds.), Computational and clinical approaches to pattern recognition and concept formation. Hillsdale, NJ: Erlbaum. https://sites.bu.edu/steveg/files/2016/06/CohenGro1990UniRecogCodes.pdf Grossberg, S. (2016). Cortical dynamics of figure-ground separation in response to 2D pictures and 3D scenes: How V2 combines border ownership, stereoscopic cues, and Gestalt grouping rules. Frontiers in Psychology. 26 January 2016. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2015.02054/full Grossberg, S., and Swaminathan, G. (2004). A laminar cortical model for 3D perception of slanted and curved surfaces and of 2D images: development, attention and bistability. Vision Research, 44, 1147-1187. https://sites.bu.edu/steveg/files/2016/06/GroSwa2004VR.pdf Congratulations on winning the best article of 2025 award in the magazine of the Military Operations Research Society! Our models have been deployed in software and hardware by MIT Lincoln Lab and other National Labs. We were awarded millions of dollars of competitive grants and contracts by essentially all the major funding agencies in Washington and elsewhere, including ARO, AFOSR, ONR, DARPA, NSF, NIH, Hughes Research Lab, etc. to develop them. We won millions of dollars in grants and contracts because our models work in the real world. Best again, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 3:40?PM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Steve, One can certainly claim that ?views? are ?parts.? It will be hard to dispute that. It?s the labeling of these ?parts? that?s the issue. In our Explanation First approach, we make a list of these ?parts? first and then build models based on that list. We have used this approach, for example, to identify different types of ships and aircraft from satellite images. Some standard parts for aircraft are like the fuselage, wings, tail and so on. The Explanation First approach is very similar to an architect understanding the needs of his/her client before designing a house. The ?Explanation First, Models Next? approach is model agnostic. So, we can use one of the ARTMAP models as the underlying model. There is no conflict there. By the way, I explained this Explanation First approach in a simple article in Phalanx last year, the magazine of the Military Operations Research Society (Home - Military Operations Research Society) and it was awarded the best article of 2025. We are starting to use the approach in certain military applications. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen > Sent: Saturday, August 29, 2026 6:25 AM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Asim, Models of the ARTMAP family (fuzzy ARTMAP, distributed ARTMAP, Gaussian ARTMAP, etc.) are perhaps the simplest kind of self-organizing neural model that can use a teacher (typically the environment, but it can also be a human) to learn to recognize objects, their parts, and their names. The work that my colleagues and I have published has gone far beyond ARTMAP. We have, in a series of articles that can be downloaded from sites.bu.edu/steveg, shown how humans can learn view-, position-, and size-invariant object representations, and use them to carry out actions that achieve valued goals. Our models show how both view-specific and view-invariant representations of an object (say, a particular view of your mother?s face AND an invariant one; namely, that she has a face) can simultaneously guide recognition of mom AND specific actions in response to where the view of her face is looking. The ?views? are ?parts?. I will list some of the relevant article titles and their urls here, for your convenience. Later articles tend to model ever more challenging aspects of self-organizing object recognition and action. DARPA gave us large grants over the years to do this work. Carpenter, G.A., Grossberg, S., and Mehanian, C. (1989). Invariant recognition of cluttered scenes by a self- organizing ART architecture: CORT-X boundary segmentation. Neural Networks, 2, 169-181. https://sites.bu.edu/steveg/files/2016/06/CarGroMeh1989NN.pdf Grossberg, S. and Wyse, L. (1991). Invariant recognition of cluttered scenes by a self-organizing ART architecture: Figure-ground separation. Neural Networks, 1991, 4, 723-742. https://sites.bu.edu/steveg/files/2016/06/GroWyse1991NN.pdf Fazl, A., Grossberg, S., and Mingolla, E. (2009). View-invariant object category learning, recognition, and search: How spatial and object attention are coordinated using surface-based attentional shrouds. Cognitive Psychology, 58, 1-48. https://sites.bu.edu/steveg/files/2016/06/FazGroMin2008.pdf Grossberg, S., and Vladusich, T. (2010). How do children learn to follow gaze, share joint attention, imitate their teachers, and use tools during social interactions? Neural Networks, 23, 940-965. https://sites.bu.edu/steveg/files/2016/06/GrossbergVladusichNN2010.pdf Cao, Y., Grossberg, S., and Markowitz, J. (2011). How does the brain rapidly learn and reorganize view- and positionally-invariant object representations in inferior temporal cortex? Neural Networks, 24, 1050-1061. https://sites.bu.edu/steveg/files/2016/06/NN2853.pdf Grossberg, S., Markowitz, J., and Cao, Y. (2011). On the road to invariant recognition: Explaining tradeoff and morph properties of cells in inferotemporal cortex using multiple-scale task-sensitive attentive learning. Neural Networks, 24, 1036-1049. https://sites.bu.edu/steveg/files/2016/06/GroMarCao2011TR.pdf Grossberg, S., Srinivasan, K., and Yazdabakhsh, A. (2011). On the road to invariant object recognition: How cortical area V2 transforms absolute to relative disparity during 3D vision. Neural Networks, 24, 686-692. https://sites.bu.edu/steveg/files/2016/06/GroSriYaz2011TR.pdf Chang, H.-C., Grossberg, S., and Cao, Y. (2014) Where?s Waldo? How perceptual cognitive, and emotional brain processes cooperate during learning to categorize and find desired objects in a cluttered scene. Frontiers in Integrative Neuroscience, doi: 10.3389/fnint.2014.0043, https://www.frontiersin.org/journals/integrative-neuroscience/articles/10.3389/fnint.2014.00043/full Grossberg, S., Srinivasan, K., and Yazdanbakhsh, A. (2014). Binocular fusion and invariant category learning due to predictive remapping during scanning of a depthful scene with eye movements. Frontiers in Psychology: Perception Science, doi: 10.3389/fpsyg.2014.01457. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.01457/full Grossberg, S. (2019). The embodied brain of SOVEREIGN2: From space-variant conscious percepts during visual search and navigation to learning invariant object categories and cognitive-emotional plans for acquiring valued goals. Frontiers in Computational Neuroscience. Published online: June 25, 2019. https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2019.00036/full Best, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 5:09?AM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Steve, What DARPA seemed to want at that time for computer vision/object recognition is object prediction based on part verification ? if a cat, have you verified its unique parts? That?s the kind of explanation they wanted. ?Teaching? an object recognition system about these parts (to generalize, parts are just subconcepts) is not inconsistent with the way we teach humans. For example, a one- or two-year-old recognizes a car. But we teach them about parts much later - such as what a tire is, a door, steering wheel and so on. How a person?s internal biological model is adjusted from these ?part teachings? I am not sure, but the identification and teaching (labeling) about parts is very much a part of human learning. One might say that LLMs automate this type of learning, but we do provide labels in the descriptions of objects. So, that?s ?teaching.? When we started with multilayered models, the assumption was that at higher levels, there would be part abstractions such as those shown in the figure below - the dog face, legs, tail and so on. The problem was, we couldn?t find these abstractions at higher levels of the network. Even if you found them, we would need to assign labels or names to them ? a dog face, legs, tail and so on. Once you get to assigning these labels, that?s ?teaching.? There is a body of work in computer vision trying to build these abstractions inside CNN models. In our method, we also ?teach? our deep learning models, but in a different way. In general, ?teaching? would be necessary whichever way you do it. We call our approach ?Explanation First, Model Next.? We have reversed the process of model building. Our testing shows gain in overall accuracy compared to some standard object detection models. This DARPA style approach also enables resistance against adversarial attacks. So substantial benefits with the Explanation First approach. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [cid:image001.png at 01DD37D5.01A7D180] From: Grossberg, Stephen > Sent: Friday, August 28, 2026 8:01 AM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Asim, Thanks for bringing up the DARPA Explainable AI program. I was invited to submit an article to a Special Issue about Explainable AI that was published when the DARPA program was announced. The editors of the Special Issue knew that Adaptive Resonance Theory, or ART, models automatically discover and incrementally learn over multiple learning trials to pay attention to the predictive, or explainable, critical feature patterns that control successful predictions and are used in learning by the adaptive weights in bottom-up adaptive filters and top-down expectations. ART does this AUTOMATICALLY. You seem to have to use an external teacher to do it. Here is the article: Grossberg, S. (2020). A path towards Explainable AI and autonomous adaptive intelligence: Deep Learning, Adaptive Resonance, and models of perception, emotion, and action. Frontiers in Neurobotics, June 25, 2020. https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2020.00036/full The article?s Abstract illustrates its explanatory range [boldface mine]: * "Biological neural network models whereby brains make minds help to understand autonomous adaptive intelligence. This article summarizes why the dynamics and emergent properties of such models for perception, cognition, emotion, and action are explainable, and thus amenable to being confidently implemented in large-scale applications. Key to their explainability is how these models combine fast activations, or short-term memory (STM) traces, and learned weights, or long-term memory (LTM) traces. Visual and auditory perceptual models have explainable conscious STM representations of visual surfaces and auditory streams in surface-shroud resonances and stream-shroud resonances, respectively. Deep Learning is often used to classify data. However, Deep Learning can experience catastrophic forgetting: At any stage of learning, an unpredictable part of its memory can collapse. Even if it makes some accurate classifications, they are not explainable and thus cannot be used with confidence. Deep Learning shares these problems with the back propagation algorithm, whose computational problems due to non-local weight transport during mismatch learning were described in the 1980s. Deep Learning became popular after very fast computers and huge online databases became available that enabled new applications despite these problems. Adaptive Resonance Theory, or ART, algorithms overcome the computational problems of back propagation and Deep Learning. ART is a self-organizing production system that incrementally learns, using arbitrary combinations of unsupervised and supervised learning and only locally computable quantities, to rapidly classify large non-stationary databases without experiencing catastrophic forgetting. ART classifications and predictions are explainable using the attended critical feature patterns in STM on which they build. The LTM adaptive weights of the fuzzy ARTMAP algorithm induce fuzzy IF-THEN rules that explain what feature combinations predict successful outcomes. ART has been successfully used in multiple large-scale real-world applications, including remote sensing, medical database prediction, and social media data clustering. Also explainable are the MOTIVATOR model of reinforcement learning and cognitive-emotional interactions, and the VITE, DIRECT, DIVA, and SOVEREIGN models for reaching, speech production, spatial navigation, and autonomous adaptive intelligence. These biological models exemplify complementary computing, and use local laws for match learning and mismatch learning that avoid the problems of Deep Learning." Chapter 5 in my 2021 Magnum Opus CONSCIOUS MIND, RESONANT BRAIN: HOW EACH BRAIN MAKES A MIND https://www.amazon.com/dp/0190070552?lv=shuf&channelId=500&plpRedirect=mhFallback provides a self-contained and non-technical overview and synthesis of my results about ART and Explainable AI. The other chapters summarize neural network models of the main processes whereby our brains make our conscious minds in healthy individuals and clinical patients. Best, Steve Stephen Grossberg Wang Professor of Cognitive and Neural Systems Director, Center for Adaptive Systems Emeritus Professor of Mathematics & Statistics, Psychological & Brain Sciences, and Biomedical Engineering Boston University sites.bu.edu/steveg/ steve at bu.edu http://en.wikipedia.org/wiki/Stephen_Grossberg http://scholar.google.com/citations?user=3BIV70wAAAAJ&hl=en https://sites.bu.edu/steveg/files/2021/08/Grossberg-CV-8-14-21.pdf https://youtu.be/9n5AnvFur7I https://www.youtube.com/watch?v=_hBye6JQCh4 https://www.amazon.com/Conscious-Mind-Resonant-Brain-Makes/dp/0190070552 https://www.amazon.com/Your-Creative-Brain-Consciously-Experience/dp/0198965370 From: Connectionists > on behalf of Asim Roy > Date: Friday, August 28, 2026 at 3:51?AM To: Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: AI at 50 videos... Stephen, You raise important issues. There are many ways of dealing with them. Here?s one of the ways. The image below shows what DARPA wanted for Explainable AI. Essentially, find the parts of an object for its prediction. So, for a cat, verify some of its unique parts like the fur, whiskers, and claws. Thus, in our explainable models, we ?teach? an object detection model about these parts. What we get as outputs, for a cat, are predictions for the cat and its parts. There?s a layer of logic that uses these outputs and verifies the existence of the parts before ?finally? predicting that there?s a cat in the image. In essence, it?s neuro-symbolic. One can do the symbolic part with a graphical model too. We can also predict that there?s a cat even if we don?t ?see? all of the parts, such as when we just see its face and not the rest of the body. And, in a similar way, your ?morning coffee? concept can be built from its parts that you mention. There?s a long tradition in computer vision of finding parts of objects inside the model. Our initial conception was that the higher-level filters in a CNN correspond to certain abstractions such as a nose, eyes, ears for a human. That was the distributed representation but based on lower-level abstractions (parts of objects). But years of research showed that one could not find these kinds of lower-level abstractions in standard CNN models. So, what do you do? Well, one way is to force some of the filters to correspond to these parts. There?s a long history to this body of work in computer vision, including that of Hinton, to create abstractions within a model. Both ways, we are ?teaching? the models about ?low-level abstraction.? In a way, in both ways, it?s a hierarchical system. In both ways, we create a neuro-symbolic system. By the way, abstractions are generalizations, they don?t code specific episodic memories. Your ?table,? ?coffee,? ?cup? and all that are abstract notions. Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [A screenshot of a computer AI-generated content may be incorrect.] From: Stephen Jos? Hanson > Sent: Thursday, August 27, 2026 4:26 AM To: Asim Roy >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Asim, I sit here with my coffee neuron and my cup neuron creating a cup of coffee, and my table neuron holding the cup and coffee neurons representing a cup of coffee. On the table, on the floor, noting of course the cup of coffee is not on the floor but on the table. So this is a network of coffee, cups, tables, maybe coded by another neuron in a hierarchical network called morning coffee? Well this is the typical problem with localist model of the brain. How would it work exactly? So far just representing a single episodic event requires hierarchical network of neurons that generalize to what exactly? Do we need many such networks to represent all episodic memories of having a cup of coffee? This is similar to the problem with the socalled "fusiform face area" which with some proper controls doesn't really exist (cf Hanson 2022). What would a small blueberry are in fusiform gyrus be doing with all those faces we recognize? A brain face cloud server? Other accounts show that FG, LO, IN and PFC are at least involved in a network of areas (not neurons) in recognizing a face. But computationally what's the actual underlying functions? The shift to "neurons represent symbols' is not helpful to explain meaning. Remember, we have billions of neurons and 100 trillions connections between them. I'm not saying symbols aren't important and Harnad's Grounding arguments are becoming more critical (and need to be made more explicit) as LLMs start to be embedded in robots. Therefore an arguably simpler task, is to identify "symbols" in LLMs since we have no idea how they work either. Are symbols in their "neurons"? Cheers Stephen On 8/24/26 22:56, Asim Roy wrote: 1. Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? 2. Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? 3. Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195?223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -- [cid:image003.jpg at 01DD37D5.01A7D180] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image001.png Type: image/png Size: 333780 bytes Desc: image001.png URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image002.jpg Type: image/jpeg Size: 45946 bytes Desc: image002.jpg URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image003.jpg Type: image/jpeg Size: 57383 bytes Desc: image003.jpg URL: From ASIM.ROY at asu.edu Sun Aug 30 18:13:10 2026 From: ASIM.ROY at asu.edu (Asim Roy) Date: Sun, 30 Aug 2026 22:13:10 +0000 Subject: Connectionists: AI@50 videos...Specialized applications vs. general-purpose self-organizing neural networks for autonomous adaptive intelligence In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> <11a8e228-8785-4b71-a2e4-d6c341d13ebe@rubic.rutgers.edu> Message-ID: Dear Steve, As I mentioned earlier, the approach is neurosymbolic. We ?teach? a neuro model about the ?symbols? we are interested in. It?s as simple as that. And we then process the symbolic output in the symbol processing unit. The symbols might be cat and its parts. Or airplanes and ships and their parts. Or drones and their parts. It?s fairly general idea and we can tie it down to how humans are taught to recognize objects. We have not tried the approach in other domains, but, in theory, it should be applicable. We do break down speech and sound into parts. And sentences into words and then syllables. Part-based recognition is powerful. It can handle adversarial attacks quite well. I would consider adversarial attacks as an extension of the concept of invariance. Our approach is model agnostic in the sense that we can as well use your ART family of models as the ?neuro? engine. Right now, we simply use fairly standard off-the-shelf object detection models. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen Sent: Sunday, August 30, 2026 7:37 AM To: Asim Roy ; Stephen Jos? Hanson ; Rothganger, Fred ; Stevan Harnad Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen Subject: Re: Connectionists: AI at 50 videos...Specialized applications vs. general-purpose self-organizing neural networks for autonomous adaptive intelligence Dear Asim, I totally agree that, if you have developed an application that works, use it! Over the years, I and my colleagues have done the same. We have regularly abstracted from our biological neural network models to design fast algorithms for efficient operation in specific application domains. E.g., Srinivasa, N., Bhattacharyya, R., Sundareswara, R., Lee, C., and Grossberg, S. (2012) A bio-inspired kinematic controller for obstacle avoidance during reaching tasks with real robots. Neural Networks, 35, 54-69. https://sites.bu.edu/steveg/files/2016/06/NN_Final.pdf Ivey, R., Bullock, D., and Grossberg, S. (2011). A neuromorphic model of spatial lookahead planning. Neural Networks, 24, 257-266. https://sites.bu.edu/steveg/files/2016/06/IveBulGro2011.pdf Grossberg, S., & Huang, T.-R. (2009). ARTSCENE: A neural system for natural scene classification. Journal of Vision, 9(4):6, 1-19. https://jov.arvojournals.org/article.aspx?articleid=2193487 Hong, S. and Grossberg, S. (2004). A neuromorphic model for achromatic and chromatic surface representation of natural images. Neural Networks, 2004, 17, 787-808. https://sites.bu.edu/steveg/files/2016/06/HongGro2004NN.pdf There are at least two possible problems with such applications: They tend to be domain-specific and, if you try to use them in a different task environment, you often hit a Brick Wall. In my modeling work during the past 69 years of how our brains make our conscious minds, I have never hit a Brick Wall. In fact, earlier principles, mechanisms, and architectures have always formed a foundation for the next discoveries. Indeed, ALL of my models can be expressed in terms of specializations of a few basic equations, and a somewhat larger number of modules or microcircuits, that are assembled into what I call MODAL architectures (for different MODALities), such as vision and image processing; audition, speech, language, and music; attention, categorization, cognitive information processing, and prediction; motivation and cognitive-emotional interactions; adaptive sensory-motor control, including reaching, standing, walking, running, tool use, dancing; and so on. That is why I am still discovering and developing new models, as my web page sites.bu.edu/steveg illustrates. It is also possible, if your current implementations do not achieve all their goals, or the goals change, for you to turn to our models for concepts and mechanisms that you can specialize for your own applications. More generally, taken together, the almost 600 downloadable archival articles on my web page provide a blueprint for the revolutionary computational paradigm of autonomous adaptive intelligence. Self-driving cars are just one example that is currently being developed. In the next century, particularly as societies and technologies need to rapidly adapt to new economic, political, and environmental challenges, increasing autonomous, adaptively intelligent solutions will help us to survive, or at least that is my hope. Best, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 8:32?PM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Steve, I am fully aware that the parts-based explanation has limits. For example, you can?t define the parts of a rock or a boulder easily, if at all, unless you bring in texture, color, and all of that. Part-based explanation is based on human understanding of objects in terms of parts and is consistent with ?teaching? humans about parts of objects. The general idea is to engage with model users and understand which parts they want to verify if a parts-based approach is acceptable for explanation. From our experience, one of the advantages of the part-based approach is that it can in fact deal with occlusions, camouflage, noise, and other kinds of distortions. For example, with this approach, if you just see the face of a cat and the body is occluded (behind a door), you can still recognize the object as a cat and justify with the parts-based logic. And the model can do that even though it was never trained on such images of cats. One can think of the parts-based approach extending the concept of invariance. We can deal with variety. For example, small drones come in different shapes and sizes, and we can do part-based recognition of flying drones. I think we all are aware of your wide body of work, which is extensive. However, I look more closely at the computer vision literature, and they directly address the problem of recognizing parts of objects within their models. Thus, my approach aligns with that body of work. And it?s consistent with the way humans learn and explain. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen > Sent: Saturday, August 29, 2026 4:24 PM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Asim, You write: "make a list of these ?parts? first and then build models based on that list.? This approach was shown decades ago to be unable to deal with novel shapes and their parts. Irv Biederman is perhaps the most distinguished psychologist to have promoted it; e.g., in his 1987 Psychological Review article (94, 115-147) entitled: Recognition-by-Components: A Theory of Human Image Understanding His approach was shown to have only limited explanatory power, since the variety of POSSIBLE novel two-dimensional and three-dimensional parts and wholes exceeds the explanatory range of the kinds of ?models? that you use. Natural objects in the real world cannot easily, if at all, be described by your approach. The speed with which you replied to my extensive list of article titles and abstracts leads me to think that you did not have enough time to even scan them. They offer rigorous computational solutions to the problem that you have posed. Please take the time to read at least the article titles and abstracts. Your approach says nothing about how a two-dimensional picture of a shape is consciously seen as the three-dimensional representation of the shape. The Necker Cube is one of hundreds of examples. Your approach also says nothing about how perceptual grouping binds distributed features into parts or wholes, depending upon Gestalt grouping laws, or how figure-ground separation occurs; notably, how we consciously see the unoccluded parts of a partly occluded object, but often only recognize, without seeing, the occluded parts. And so on... Here are a few additional relevant articles: Cohen, M.A. and Grossberg, S. (1990). Unitized recognition codes for parts and wholes: The unique cue in configural discriminations. In M. Commons, R. Herrnstein, S. Kosslyn, and D. Mumford (Eds.), Computational and clinical approaches to pattern recognition and concept formation. Hillsdale, NJ: Erlbaum. https://sites.bu.edu/steveg/files/2016/06/CohenGro1990UniRecogCodes.pdf Grossberg, S. (2016). Cortical dynamics of figure-ground separation in response to 2D pictures and 3D scenes: How V2 combines border ownership, stereoscopic cues, and Gestalt grouping rules. Frontiers in Psychology. 26 January 2016. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2015.02054/full Grossberg, S., and Swaminathan, G. (2004). A laminar cortical model for 3D perception of slanted and curved surfaces and of 2D images: development, attention and bistability. Vision Research, 44, 1147-1187. https://sites.bu.edu/steveg/files/2016/06/GroSwa2004VR.pdf Congratulations on winning the best article of 2025 award in the magazine of the Military Operations Research Society! Our models have been deployed in software and hardware by MIT Lincoln Lab and other National Labs. We were awarded millions of dollars of competitive grants and contracts by essentially all the major funding agencies in Washington and elsewhere, including ARO, AFOSR, ONR, DARPA, NSF, NIH, Hughes Research Lab, etc. to develop them. We won millions of dollars in grants and contracts because our models work in the real world. Best again, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 3:40?PM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Steve, One can certainly claim that ?views? are ?parts.? It will be hard to dispute that. It?s the labeling of these ?parts? that?s the issue. In our Explanation First approach, we make a list of these ?parts? first and then build models based on that list. We have used this approach, for example, to identify different types of ships and aircraft from satellite images. Some standard parts for aircraft are like the fuselage, wings, tail and so on. The Explanation First approach is very similar to an architect understanding the needs of his/her client before designing a house. The ?Explanation First, Models Next? approach is model agnostic. So, we can use one of the ARTMAP models as the underlying model. There is no conflict there. By the way, I explained this Explanation First approach in a simple article in Phalanx last year, the magazine of the Military Operations Research Society (Home - Military Operations Research Society) and it was awarded the best article of 2025. We are starting to use the approach in certain military applications. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) From: Grossberg, Stephen > Sent: Saturday, August 29, 2026 6:25 AM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Self-organized learning of objects and their parts Dear Asim, Models of the ARTMAP family (fuzzy ARTMAP, distributed ARTMAP, Gaussian ARTMAP, etc.) are perhaps the simplest kind of self-organizing neural model that can use a teacher (typically the environment, but it can also be a human) to learn to recognize objects, their parts, and their names. The work that my colleagues and I have published has gone far beyond ARTMAP. We have, in a series of articles that can be downloaded from sites.bu.edu/steveg, shown how humans can learn view-, position-, and size-invariant object representations, and use them to carry out actions that achieve valued goals. Our models show how both view-specific and view-invariant representations of an object (say, a particular view of your mother?s face AND an invariant one; namely, that she has a face) can simultaneously guide recognition of mom AND specific actions in response to where the view of her face is looking. The ?views? are ?parts?. I will list some of the relevant article titles and their urls here, for your convenience. Later articles tend to model ever more challenging aspects of self-organizing object recognition and action. DARPA gave us large grants over the years to do this work. Carpenter, G.A., Grossberg, S., and Mehanian, C. (1989). Invariant recognition of cluttered scenes by a self- organizing ART architecture: CORT-X boundary segmentation. Neural Networks, 2, 169-181. https://sites.bu.edu/steveg/files/2016/06/CarGroMeh1989NN.pdf Grossberg, S. and Wyse, L. (1991). Invariant recognition of cluttered scenes by a self-organizing ART architecture: Figure-ground separation. Neural Networks, 1991, 4, 723-742. https://sites.bu.edu/steveg/files/2016/06/GroWyse1991NN.pdf Fazl, A., Grossberg, S., and Mingolla, E. (2009). View-invariant object category learning, recognition, and search: How spatial and object attention are coordinated using surface-based attentional shrouds. Cognitive Psychology, 58, 1-48. https://sites.bu.edu/steveg/files/2016/06/FazGroMin2008.pdf Grossberg, S., and Vladusich, T. (2010). How do children learn to follow gaze, share joint attention, imitate their teachers, and use tools during social interactions? Neural Networks, 23, 940-965. https://sites.bu.edu/steveg/files/2016/06/GrossbergVladusichNN2010.pdf Cao, Y., Grossberg, S., and Markowitz, J. (2011). How does the brain rapidly learn and reorganize view- and positionally-invariant object representations in inferior temporal cortex? Neural Networks, 24, 1050-1061. https://sites.bu.edu/steveg/files/2016/06/NN2853.pdf Grossberg, S., Markowitz, J., and Cao, Y. (2011). On the road to invariant recognition: Explaining tradeoff and morph properties of cells in inferotemporal cortex using multiple-scale task-sensitive attentive learning. Neural Networks, 24, 1036-1049. https://sites.bu.edu/steveg/files/2016/06/GroMarCao2011TR.pdf Grossberg, S., Srinivasan, K., and Yazdabakhsh, A. (2011). On the road to invariant object recognition: How cortical area V2 transforms absolute to relative disparity during 3D vision. Neural Networks, 24, 686-692. https://sites.bu.edu/steveg/files/2016/06/GroSriYaz2011TR.pdf Chang, H.-C., Grossberg, S., and Cao, Y. (2014) Where?s Waldo? How perceptual cognitive, and emotional brain processes cooperate during learning to categorize and find desired objects in a cluttered scene. Frontiers in Integrative Neuroscience, doi: 10.3389/fnint.2014.0043, https://www.frontiersin.org/journals/integrative-neuroscience/articles/10.3389/fnint.2014.00043/full Grossberg, S., Srinivasan, K., and Yazdanbakhsh, A. (2014). Binocular fusion and invariant category learning due to predictive remapping during scanning of a depthful scene with eye movements. Frontiers in Psychology: Perception Science, doi: 10.3389/fpsyg.2014.01457. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.01457/full Grossberg, S. (2019). The embodied brain of SOVEREIGN2: From space-variant conscious percepts during visual search and navigation to learning invariant object categories and cognitive-emotional plans for acquiring valued goals. Frontiers in Computational Neuroscience. Published online: June 25, 2019. https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2019.00036/full Best, Steve Get Outlook for Mac From: Asim Roy > Date: Saturday, August 29, 2026 at 5:09?AM To: Grossberg, Stephen >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: RE: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Steve, What DARPA seemed to want at that time for computer vision/object recognition is object prediction based on part verification ? if a cat, have you verified its unique parts? That?s the kind of explanation they wanted. ?Teaching? an object recognition system about these parts (to generalize, parts are just subconcepts) is not inconsistent with the way we teach humans. For example, a one- or two-year-old recognizes a car. But we teach them about parts much later - such as what a tire is, a door, steering wheel and so on. How a person?s internal biological model is adjusted from these ?part teachings? I am not sure, but the identification and teaching (labeling) about parts is very much a part of human learning. One might say that LLMs automate this type of learning, but we do provide labels in the descriptions of objects. So, that?s ?teaching.? When we started with multilayered models, the assumption was that at higher levels, there would be part abstractions such as those shown in the figure below - the dog face, legs, tail and so on. The problem was, we couldn?t find these abstractions at higher levels of the network. Even if you found them, we would need to assign labels or names to them ? a dog face, legs, tail and so on. Once you get to assigning these labels, that?s ?teaching.? There is a body of work in computer vision trying to build these abstractions inside CNN models. In our method, we also ?teach? our deep learning models, but in a different way. In general, ?teaching? would be necessary whichever way you do it. We call our approach ?Explanation First, Model Next.? We have reversed the process of model building. Our testing shows gain in overall accuracy compared to some standard object detection models. This DARPA style approach also enables resistance against adversarial attacks. So substantial benefits with the Explanation First approach. Best, Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [cid:image001.png at 01DD388F.878E4670] From: Grossberg, Stephen > Sent: Friday, August 28, 2026 8:01 AM To: Asim Roy >; Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu; Grossberg, Stephen > Subject: Re: Connectionists: AI at 50 videos...Explainable AI emerges naturally from Adaptive Resonance Theory Dear Asim, Thanks for bringing up the DARPA Explainable AI program. I was invited to submit an article to a Special Issue about Explainable AI that was published when the DARPA program was announced. The editors of the Special Issue knew that Adaptive Resonance Theory, or ART, models automatically discover and incrementally learn over multiple learning trials to pay attention to the predictive, or explainable, critical feature patterns that control successful predictions and are used in learning by the adaptive weights in bottom-up adaptive filters and top-down expectations. ART does this AUTOMATICALLY. You seem to have to use an external teacher to do it. Here is the article: Grossberg, S. (2020). A path towards Explainable AI and autonomous adaptive intelligence: Deep Learning, Adaptive Resonance, and models of perception, emotion, and action. Frontiers in Neurobotics, June 25, 2020. https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2020.00036/full The article?s Abstract illustrates its explanatory range [boldface mine]: * "Biological neural network models whereby brains make minds help to understand autonomous adaptive intelligence. This article summarizes why the dynamics and emergent properties of such models for perception, cognition, emotion, and action are explainable, and thus amenable to being confidently implemented in large-scale applications. Key to their explainability is how these models combine fast activations, or short-term memory (STM) traces, and learned weights, or long-term memory (LTM) traces. Visual and auditory perceptual models have explainable conscious STM representations of visual surfaces and auditory streams in surface-shroud resonances and stream-shroud resonances, respectively. Deep Learning is often used to classify data. However, Deep Learning can experience catastrophic forgetting: At any stage of learning, an unpredictable part of its memory can collapse. Even if it makes some accurate classifications, they are not explainable and thus cannot be used with confidence. Deep Learning shares these problems with the back propagation algorithm, whose computational problems due to non-local weight transport during mismatch learning were described in the 1980s. Deep Learning became popular after very fast computers and huge online databases became available that enabled new applications despite these problems. Adaptive Resonance Theory, or ART, algorithms overcome the computational problems of back propagation and Deep Learning. ART is a self-organizing production system that incrementally learns, using arbitrary combinations of unsupervised and supervised learning and only locally computable quantities, to rapidly classify large non-stationary databases without experiencing catastrophic forgetting. ART classifications and predictions are explainable using the attended critical feature patterns in STM on which they build. The LTM adaptive weights of the fuzzy ARTMAP algorithm induce fuzzy IF-THEN rules that explain what feature combinations predict successful outcomes. ART has been successfully used in multiple large-scale real-world applications, including remote sensing, medical database prediction, and social media data clustering. Also explainable are the MOTIVATOR model of reinforcement learning and cognitive-emotional interactions, and the VITE, DIRECT, DIVA, and SOVEREIGN models for reaching, speech production, spatial navigation, and autonomous adaptive intelligence. These biological models exemplify complementary computing, and use local laws for match learning and mismatch learning that avoid the problems of Deep Learning." Chapter 5 in my 2021 Magnum Opus CONSCIOUS MIND, RESONANT BRAIN: HOW EACH BRAIN MAKES A MIND https://www.amazon.com/dp/0190070552?lv=shuf&channelId=500&plpRedirect=mhFallback provides a self-contained and non-technical overview and synthesis of my results about ART and Explainable AI. The other chapters summarize neural network models of the main processes whereby our brains make our conscious minds in healthy individuals and clinical patients. Best, Steve Stephen Grossberg Wang Professor of Cognitive and Neural Systems Director, Center for Adaptive Systems Emeritus Professor of Mathematics & Statistics, Psychological & Brain Sciences, and Biomedical Engineering Boston University sites.bu.edu/steveg/ steve at bu.edu http://en.wikipedia.org/wiki/Stephen_Grossberg http://scholar.google.com/citations?user=3BIV70wAAAAJ&hl=en https://sites.bu.edu/steveg/files/2021/08/Grossberg-CV-8-14-21.pdf https://youtu.be/9n5AnvFur7I https://www.youtube.com/watch?v=_hBye6JQCh4 https://www.amazon.com/Conscious-Mind-Resonant-Brain-Makes/dp/0190070552 https://www.amazon.com/Your-Creative-Brain-Consciously-Experience/dp/0198965370 From: Connectionists > on behalf of Asim Roy > Date: Friday, August 28, 2026 at 3:51?AM To: Stephen Jos? Hanson >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu > Subject: Re: Connectionists: AI at 50 videos... Stephen, You raise important issues. There are many ways of dealing with them. Here?s one of the ways. The image below shows what DARPA wanted for Explainable AI. Essentially, find the parts of an object for its prediction. So, for a cat, verify some of its unique parts like the fur, whiskers, and claws. Thus, in our explainable models, we ?teach? an object detection model about these parts. What we get as outputs, for a cat, are predictions for the cat and its parts. There?s a layer of logic that uses these outputs and verifies the existence of the parts before ?finally? predicting that there?s a cat in the image. In essence, it?s neuro-symbolic. One can do the symbolic part with a graphical model too. We can also predict that there?s a cat even if we don?t ?see? all of the parts, such as when we just see its face and not the rest of the body. And, in a similar way, your ?morning coffee? concept can be built from its parts that you mention. There?s a long tradition in computer vision of finding parts of objects inside the model. Our initial conception was that the higher-level filters in a CNN correspond to certain abstractions such as a nose, eyes, ears for a human. That was the distributed representation but based on lower-level abstractions (parts of objects). But years of research showed that one could not find these kinds of lower-level abstractions in standard CNN models. So, what do you do? Well, one way is to force some of the filters to correspond to these parts. There?s a long history to this body of work in computer vision, including that of Hinton, to create abstractions within a model. Both ways, we are ?teaching? the models about ?low-level abstraction.? In a way, in both ways, it?s a hierarchical system. In both ways, we create a neuro-symbolic system. By the way, abstractions are generalizations, they don?t code specific episodic memories. Your ?table,? ?coffee,? ?cup? and all that are abstract notions. Asim Asim Roy Professor, Information Systems Arizona State University Asim Roy | iSearch (asu.edu) [A screenshot of a computer AI-generated content may be incorrect.] From: Stephen Jos? Hanson > Sent: Thursday, August 27, 2026 4:26 AM To: Asim Roy >; Rothganger, Fred >; Stevan Harnad > Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... Asim, I sit here with my coffee neuron and my cup neuron creating a cup of coffee, and my table neuron holding the cup and coffee neurons representing a cup of coffee. On the table, on the floor, noting of course the cup of coffee is not on the floor but on the table. So this is a network of coffee, cups, tables, maybe coded by another neuron in a hierarchical network called morning coffee? Well this is the typical problem with localist model of the brain. How would it work exactly? So far just representing a single episodic event requires hierarchical network of neurons that generalize to what exactly? Do we need many such networks to represent all episodic memories of having a cup of coffee? This is similar to the problem with the socalled "fusiform face area" which with some proper controls doesn't really exist (cf Hanson 2022). What would a small blueberry are in fusiform gyrus be doing with all those faces we recognize? A brain face cloud server? Other accounts show that FG, LO, IN and PFC are at least involved in a network of areas (not neurons) in recognizing a face. But computationally what's the actual underlying functions? The shift to "neurons represent symbols' is not helpful to explain meaning. Remember, we have billions of neurons and 100 trillions connections between them. I'm not saying symbols aren't important and Harnad's Grounding arguments are becoming more critical (and need to be made more explicit) as LLMs start to be embedded in robots. Therefore an arguably simpler task, is to identify "symbols" in LLMs since we have no idea how they work either. Are symbols in their "neurons"? Cheers Stephen On 8/24/26 22:56, Asim Roy wrote: 1. Plate (2002): ?Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).? 2. Thorpe (1995, p. 550): ?With a local representation, activity in individual units can be interpreted directly ? with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.? 3. Elman (1995, p. 210): ?These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.? 1. Elman J. (1995). Language as a dynamical system, in Mind as Motion: Explorations in the Dynamics of Cognition, eds Port R., van Gelder T. (Cambridge, MA: MIT Press), 195?223. 2. PlateT. (2002). Distributed representations, in: Encyclopedia of Cognitive Science, ed Nadel L. (London: Macmillan), 2. 3. ThorpeS. (1995). Localized versus distributed representations, in The Handbook of Brain Theory and Neural Networks, ed Arbib M. (Cambridge, MA: MIT Press), 550. In that discussion with Horace Barlow and others, interpretation and meaning of the activations that responded only to certain specific stimuli was the main source of discomfort to connectionists including Walter Freeman. Hence Walter asking Rodrigo directly whether those activations had meaning and interpretation. Asim Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) From: Rothganger, Fred Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and ?meaning.? Reddy and Thorpe (2014) conclude: ?In conclusion, evidence is accumulating that ?concept cells? carry high-level, abstract stimulus information.? Single cells having ?meaning? was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had ?meaning.? ?Meaning? and abstraction is at the heart of these debates, not grandmother cells. And finding ?meaning? in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -- [cid:image003.jpg at 01DD388F.878E4670] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image001.png Type: image/png Size: 333780 bytes Desc: image001.png URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image002.jpg Type: image/jpeg Size: 45946 bytes Desc: image002.jpg URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image003.jpg Type: image/jpeg Size: 57383 bytes Desc: image003.jpg URL: From suashdeb at gmail.com Mon Aug 31 06:09:35 2026 From: suashdeb at gmail.com (Suash Deb) Date: Mon, 31 Aug 2026 15:39:35 +0530 Subject: Connectionists: Final extn of deadline, ISCMI 2026 (Vienna) Message-ID: Dear Friends and Esteemed Colleagues, Warm greetings. Trust you are doing well. This is to share with you that on numerous requests, the deadline for submissions for ISCMI 2026 , the annual flagship event of llCCl and technically co sponsored by IEEE Austria Section has been extended till 30th September 2026 http://www.iscmi.us Pls note that this is the final deadline and no further extension will be granted for this conference. Hope this will facilitate submissions for the peers, who could not complete their manuscripts but intend to submit for ISCMI26 . An updated leaflet of the conference is . Will look forward to receiving more submissions from you/your peers in the coming days and with best regards, Suash Deb General Chair, ISCMI 2024 -------------- next part -------------- An HTML attachment was scrubbed... URL: From alessandro.gifford at gmail.com Mon Aug 31 09:06:33 2026 From: alessandro.gifford at gmail.com (Alessandro Gifford) Date: Mon, 31 Aug 2026 15:06:33 +0200 Subject: Connectionists: Algonauts Project 2027 Challenge Message-ID: Dear colleagues, We are excited to announce the Algonauts Project Challenge 2027 , organized in collaboration with CNeuroMod : How The Human Brain Makes Sense of Video Games. The goal of this year?s challenge is to predict human brain responses measured with fMRI during a variety of video games. Moving beyond passive stimulus viewing, the challenge brings together perception, decision-making, behavior, and learning, pushing brain modeling toward the complex and dynamic environments in which human cognition naturally unfolds. The challenge will launch at the end of 2026 and run until summer 2027. The top three teams will be invited to present their work at the Algonauts Challenge showcase at CCN 2027 in Edinburgh. More details about the challenge will follow over the coming months. To make sure you do not miss any updates, sign up here . Website: https://algonautsproject.com/2027/index.html Email: algonauts.mit at gmail.com Bluesky: @algonautsproject.bsky.social X: @AlgonautsProj We look forward to your participation in the challenge! Best, The Algonauts Project Team Domenic Bersch, Goethe University Frankfurt Alessandro Gifford, Freie Universit?t Berlin Marie St-Laurent, CRIUGM Basile Pinsard, CRIUGM Yann Harel, Boston College Julie Boyle, Universit? de Montr?al Lune Bellec, Universit? de Montr?al Aude Oliva, Massachusetts Institute of Technology Gemma Roig, Goethe University Frankfurt Radoslaw Cichy, Freie Universit?t Berlin -------------- next part -------------- An HTML attachment was scrubbed... URL: From stefan.wermter at uni-hamburg.de Mon Aug 31 10:59:00 2026 From: stefan.wermter at uni-hamburg.de (Stefan Wermter) Date: Mon, 31 Aug 2026 16:59:00 +0200 Subject: Connectionists: AI@50, AI@70 and towards continual embodied learning Message-ID: Dear all I enjoyed our discussion on AI at 50 and beyond. Lots of exciting progress on neural and hybrid neural symbolic research since AI at 50, AI at 70 and now going beyond. In the future,?embodied neural intelligence based on?continual developmental learning may play an even more central role since the body and mind are highly integrated in humans right from the beginning. New paper in this context from last week "Robot in a crib: How a playing robot helps us understand sensorimotor contingency learning" just published in Science Robotics. Authors: Josua Spisak, Sergiu T. Popescu, Luk?? Rustler, Stefan Wermter, J. Kevin O'Regan, Matej Hoffmann Nutshell: an iCub lying on its back in a crib learns to interact with its environment biologically inspired by infants. Our embodied model learns through a balanced trade-off between curiosity and prediction and manages to adjust its behavior towards specific conditions. We found that the emergent strategies do not only rely on a single metric such as activity, but ?complex and dynamic strategies emerge that make use of spatial positions, environmental dynamics, and specific movements. Limited full-text access link: https://www.science.org/eprint/KJ3AVYAICKSA5C83NVZR/full?activationRedirect=/doi/full/10.1126/scirobotics.aed4106 best Stefan -- *Professor Dr. Stefan Wermter* Director of Knowledge Technology Department of Informatics University of Hamburg Bundesstrasse 56b 20146 Hamburg, Germany stefan.wermter at uni-hamburg.de http://www.informatik.uni-hamburg.de/WTM/ -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: smime.p7s Type: application/pkcs7-signature Size: 4152 bytes Desc: S/MIME Cryptographic Signature URL: From ASIM.ROY at asu.edu Mon Aug 31 22:51:42 2026 From: ASIM.ROY at asu.edu (Asim Roy) Date: Tue, 1 Sep 2026 02:51:42 +0000 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Dan, That is the Walter I remember. Even though we had differing views, he was always ready to help, even submitting funding proposals together. Asim From: Levine, Daniel S Sent: Monday, August 31, 2026 7:48 PM To: Asim Roy ; Rothganger, Fred Cc: connectionists at mailman.srv.cs.cmu.edu Subject: RE: AI at 50 videos... Asim, >From what I remember of Walter, I can picture him arguing with a smile on his face, and not getting angry with anybody. Dan From: Connectionists > On Behalf Of Asim Roy Sent: Monday, August 24, 2026 10:17 PM To: Rothganger, Fred > Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... [External] Although we argued a lot, Walter Freeman and I were good friends. Asim From: Rothganger, Fred > Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy > Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and "meaning." Reddy and Thorpe (2014) conclude: "In conclusion, evidence is accumulating that "concept cells" carry high-level, abstract stimulus information." Single cells having "meaning" was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had "meaning." "Meaning" and abstraction is at the heart of these debates, not grandmother cells. And finding "meaning" in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -------------- next part -------------- An HTML attachment was scrubbed... URL: From levine at uta.edu Mon Aug 31 22:47:35 2026 From: levine at uta.edu (Levine, Daniel S) Date: Tue, 1 Sep 2026 02:47:35 +0000 Subject: Connectionists: AI@50 videos... In-Reply-To: References: <790AFD45-3320-4E74-A668-97E07F83F424@nyu.edu> Message-ID: Asim, >From what I remember of Walter, I can picture him arguing with a smile on his face, and not getting angry with anybody. Dan From: Connectionists On Behalf Of Asim Roy Sent: Monday, August 24, 2026 10:17 PM To: Rothganger, Fred Cc: connectionists at mailman.srv.cs.cmu.edu Subject: Re: Connectionists: AI at 50 videos... [External] Although we argued a lot, Walter Freeman and I were good friends. Asim From: Rothganger, Fred > Sent: Monday, August 24, 2026 8:46 AM To: Asim Roy > Cc: connectionists at mailman.srv.cs.cmu.edu Subject: AI at 50 videos... Asim, It makes sense for you to put "meaning" in scare-quotes, since it is another ill-defined term. IMHO, the important thing is adaptive behavior. That a particular neuron fires when the organism is exposed to a stimulus is significant because that spike does work on other parts of the system, ultimately shaping behavior. The notion of "meaning" may just be a convenience for us as observers discussing the system. -- Fred ________________________________ From: Connectionists > on behalf of Asim Roy > Sent: Friday, August 21, 2026 10:47 PM To: KENTRIDGE, ROBERT W. >; Gary Marcus >; Stephen Jos? Hanson >; Hava Siegelmann (hava.siegelmann at gmail.com) >; Juergen Schmidhuber >; Ali Minai > Cc: director at inc.ucsd.edu >; connectionists at mailman.srv.cs.cmu.edu > Subject: [EXTERNAL] Re: Connectionists: FW: AI at 50 videos... By the way, most of these microelectrode-based studies, starting with Horace Barlow, are single cell studies. And the single cell studies are the ones that have won Noble prizes because of the insights they produced. The overall evidence from these studies is not just about grandmother cells, but about single cells encoding abstractions and "meaning." Reddy and Thorpe (2014) conclude: "In conclusion, evidence is accumulating that "concept cells" carry high-level, abstract stimulus information." Single cells having "meaning" was (and still is) at the heart of discussion at that time because it contradicts the fundamental population coding idea where single neurons by themselves have no meaning, they only have meaning collectively. And Walter Freeman at that time could not accept the idea that single cells have meaning. We were supposed to have a public debate about this at IJCNN 2011 in San Jose. Walter posed the question directly to Rodrigo Quian Quiroga, who conducted many of these experiments under the supervision of Itzhak Fried and Christof Koch. Walter also knew Rodrigo, having been co-authors or co-editors of a book. Rodrigo came through at the last minute before the public debate, acknowledging that the single cells in his studies indeed had "meaning." "Meaning" and abstraction is at the heart of these debates, not grandmother cells. And finding "meaning" in activations of single cells directly contradicts the population coding hypothesis and supports instead the symbol system hypothesis. Asim Roy Professor, Information Systems Arizona State University Lifeboat Foundation Bios: Professor Asim Roy Asim Roy | iSearch (asu.edu) -------------- next part -------------- An HTML attachment was scrubbed... URL: From cms at hum.ku.dk Mon Aug 31 14:22:28 2026 From: cms at hum.ku.dk (Camilo Miguel Signorelli) Date: Mon, 31 Aug 2026 18:22:28 +0000 Subject: Connectionists: Models of Consciousness 2026 - Copenhagen - Registration deadline Message-ID: <3d49350805aa43a9bb63f81e5b81aa5d@hum.ku.dk> Dear colleagues, Apologies in advance for any cross-communication. As chair of the Models of Consciousness Conference 2026 (12-16 October 2026, Copenhagen), I would like to call your attention to this year's program, including the first transdisciplinary collaborative exercise to clarify concepts and set future priorities in the field. We have announced the project and conference across several channels. Not sure whether it was shared on the connectionist list. The official registration deadline is today, 31st of August 2026. With more than 270 abstracts submitted, we are close to our maximum capacity. Late registration may be possible as long as we have spots left. Do not hesitate to reach out via the official MoC email. Abrazos grandes, PS: For news about the project during MoC, you can also follow the MoC official LinkedIn account. PhD. DPhil. Camilo Miguel Signorelli | ConScience CPAI Department of Communication University of Copenhagen Karen Blixens Plads 8 2300 K?benhavn S cms at hum.ku.dk -------------- next part -------------- An HTML attachment was scrubbed... URL: