Connectionists: CFP: Correlative Learning Workshop at IJCNN 2023

[Anderson,Chuck]

The submission deadline has been extended to April 23rd.

See https://correlativelearning.ai/ for details.

Workshop Topics
-----------------------

Despite gradient descent being the dominant approach to parameter adaptation in neural networks, correlative learning is often considered to be more biologically plausible, easier to implement in hardware and an interesting alternative to the gradient-based methods. Our workshop is a venue for the discussion of correlation-based learning methods, their applications, recent developments and future directions. This workshop will help build a community of researchers interested in discussing common problems and elucidating the most important questions facing these types of approaches.

Call for Participation
---------------------------

Correlative Learning (CL) uses product terms, rather than derivatives, to adapt the parameters of learning systems. Though CL has had significant impact on neuroscience, signal processing, and control, the Machine Learning (ML) community still predominantly uses derivative-based approaches. Several CL approaches were introduced in the 1990s and shown to be effective in Neural Networks with a few layers and a moderate number of parameters. One of the best known examples of CL is Hebbian learning. Some others, like Alopex are applied in a stochastic framework. In this workshop we plan to thoroughly investigate the effectiveness of CL for current Deep Learning (DL) systems; this could also enhance the effectiveness of other methods.

This workshop seeks to bring together practitioners of CL from Neuroscience, Psychology, Computer Science, Physics, Mathematics, and other fields. Discussion of recent results across a diverse range of applications can create an active community of CL researchers and this can lead to dissemination of CL approaches more broadly to the ML community. We plan to devote about half of the workshop to lively discussions, based on a set of questions, like: “How can we get better approximate gradient information through correlations?”, “What are the implementational advantages of CL in hardware and software?”, “How can we scale CL to deeper, wider networks?”, “Can CL benefit from adaptive momentum in the ADAM and related algorithms?”, “How can biological CL models inform ML?”. Also, an entire session of the workshop will be devoted to graduate students and post-docs, where presentations and discussions about their nascent ideas can help create the next generation CL community.

We invite anyone interested in Correlative Learning to join us at IJCNN 2023 either in person or virtually.

Submission Details
--------------------------

In order to maximise time for discussion and encourage the development of the field, we have scheduled five colloquia and six early career researcher presentations (open to graduate students and those who were awarded a PhD within the last 10 years). Colloquia and presentations will be selected by the program committee. Criteria for selection are: appropriateness to the workshop and merit. To submit a colloquium proposal send (1) your name, (2) a brief biography, (3) a title, (4) an abstract of 300 words or less (not-including references), (5) for early career a statement regarding your qualification for the category signed by a graduate advisor, to abstract at correlativelearning.ai by no later than the end of day April 23rd (AOE).




Chuck Anderson

Department of Computer Science

Colorado State University

http://www.cs.colostate.edu/~anderson
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20230420/8e816a93/attachment.html>