From arielpro at gmail.com Fri Jan 1 15:01:13 2016 From: arielpro at gmail.com (Ariel Procaccia) Date: Fri, 1 Jan 2016 15:01:13 -0500 Subject: [AI Seminar] AI lunch call for talks In-Reply-To: References: Message-ID: Hi everyone, so far we have very few volunteers. Most importantly, we'd like to fill the first slots in January: 12, 19 and 26. Please email Ellen (and cc me) if you'd be willing to give a 50 minute talk on one of these dates. Thanks, Ariel On Tue, Dec 22, 2015 at 12:54 PM, Ellen Vitercik wrote: > Hi everyone! > > Ariel and I will be organizing AI lunch beginning this spring semester. > Lunches will be held every Tuesday from noon to 1 in NSH 3305, unless we > announce otherwise. We hope to have outside speakers come about once per > month and we encourage students and post-docs to give talks during the > remaining weeks. AI lunch will be a great venue for you to present your > work to a broad AI audience, so we hope that you will help us ensure a > successful start to this new tradition by volunteering to give a talk. > > If you are interested in giving a 50 minute talk, please email me at > evitercik at cs.cmu.edu with your scheduling constraints. > > Best, > Ellen and Ariel > -------------- next part -------------- An HTML attachment was scrubbed... URL: From evitercik at cs.cmu.edu Thu Jan 7 13:08:34 2016 From: evitercik at cs.cmu.edu (Ellen Vitercik) Date: Thu, 7 Jan 2016 13:08:34 -0500 Subject: [AI Seminar] AI Lunch - Jun Zhu - January 12, 2016 Message-ID: Dear faculty and students, Please join us for this semester's first AI lunch this Tuesday, January 12th at 12pm in NSH 3305. Jun Zhu , Associate Professor of Computer Science at Tsinghua University and Adjunct Faculty of Machine Learning at Carnegie Mellon University will be giving the talk. We will send the title and talk abstract in the coming days. We will continue to update the AI Lunch and Seminar webpage as the semester progresses. If you are interested in giving a talk on any Tuesday this semester that has not yet been scheduled, please write to me and Ariel. Best, Ellen and Ariel evitercik at cs.cmu.edu arielpro at cs.cmu.edu -------------- next part -------------- An HTML attachment was scrubbed... URL: From awm at cs.cmu.edu Thu Jan 7 13:16:05 2016 From: awm at cs.cmu.edu (Andrew Moore) Date: Thu, 7 Jan 2016 13:16:05 -0500 Subject: [AI Seminar] AI Lunch - Jun Zhu - January 12, 2016 In-Reply-To: References: Message-ID: Sorry for the extra noise on this channel, but please let me say how excited I am that this tradition has been resurrected. I believe there is a huge growth spurt in AI education and research coming down the pike for SCS and so this reboot is fantastic. I would join on Jan 12th except I'm in California, ironically talking to two major companies about how AI is their key growth area in the next 5 years and why they need to team with CMU! I will eagerly attend these meetings when I am in town. On Thu, Jan 7, 2016 at 1:08 PM, Ellen Vitercik wrote: > Dear faculty and students, > > Please join us for this semester's first AI lunch this Tuesday, January > 12th at 12pm in NSH 3305. Jun Zhu > , Associate Professor > of Computer Science at Tsinghua University and Adjunct Faculty of Machine > Learning at Carnegie Mellon University will be giving the talk. We will > send the title and talk abstract in the coming days. > > We will continue to update the AI Lunch and Seminar webpage > as the semester progresses. > If you are interested in giving a talk on any Tuesday this semester that > has not yet been scheduled, please write to me and Ariel. > > Best, > > Ellen and Ariel > evitercik at cs.cmu.edu > arielpro at cs.cmu.edu > -- Andrew Moore , Dean, School of Computer Science , Carnegie Mellon. Twitter feed -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Jan 11 09:17:21 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 11 Jan 2016 09:17:21 -0500 Subject: [AI Seminar] AI Lunch - Room Change and Talk Information Message-ID: Dear faculty and students, We look forward to seeing you tomorrow, Tuesday, January 12th, at noon for the first AI lunch of the semester. There is a room change -- we will meet in *GHC 6115.* Jun Zhu will speak about Scalable Bayesian Inference with Posterior Regularization. Below is a talk abstract and speaker bio. *Abstract:* Bayesian methods represent one important type of statistical methods for learning, inference and decision making. At the core is Bayes' theorem, which has been developed for more than 250 years. However, in the Big Data era, many challenges need to be addressed, ranging from theory, algorithm, and applications. In this talk, I will introduce some recent developments on Bayesian inference with posterior regularization, which can incorporate rich side information such as the large-margin property we like to impose on the model distribution for accurate prediction, or the domain knowledge collected from experts or crowds for good interpretation, and scalable online learning and distributed inference algorithms. When applied to deep generative models, we are able to significantly improve the prediction accuracy without sacrificing the generative performance. *Bio: *Dr. Jun Zhu is an associate professor at Department of Computer Science and Technology, Tsinghua University, and an adjunct faculty at Machine Learning Department, Carnegie Mellon University. He received his Ph.D. in Computer Science from Tsinghua in 2009. Before joining Tsinghua in 2011, he did post-doctoral research in CMU. His research interest lies in developing scalable machine learning methods to understand complex scientific and engineering data. Dr. Zhu has published over 60 peer-reviewed papers in the prestigious conferences and journals. He is an associate editor for IEEE Trans. on PAMI. He served as area chair for ICML, NIPS, UAI, IJCAI and AAAI. He was a local chair of ICML 2014. He is a recipient of the IEEE Intelligent Systems "AI's 10 to Watch" Award, NSFC Excellent Young Scholar Award, and CCF Young Scientist Award. His work is supported by the "221 Basic Research Plan for Young Talents" at Tsinghua. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Thu Jan 14 16:48:28 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Thu, 14 Jan 2016 16:48:28 -0500 Subject: [AI Seminar] AI Lunch - Noam Brown - January 19, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, January 19th at noon in *NSH 3305* for AI lunch . There will be Thai food! Noam Brown will speak about Simultaneous Abstraction and Equilibrium Finding. *Abstract: *The leading approach for solving large games of imperfect information is to find a Nash equilibrium in a smaller, abstract version of the game and then map the solution back to the original game. However, it is difficult to know what information and which actions to discard or include in the abstraction without first solving the game, and it is infeasible to solve the game without first abstracting it. We introduce a method that combines abstraction with equilibrium finding by enabling the abstraction to change during run time. This allows an agent to begin learning with a coarse abstraction, and then to strategically refine the abstraction at points that the strategy computed in the current abstraction deems important. The algorithm can switch to the refined abstraction at the cost of a single traversal while provably not having to restart the equilibrium finding. Experiments show it can outperform fixed abstractions at every stage of the run: early on it improves as quickly as equilibrium finding in coarse abstractions, and later it converges to a better solution than does equilibrium finding in fine-grained abstractions. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Jan 18 10:31:37 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 18 Jan 2016 10:31:37 -0500 Subject: [AI Seminar] Fwd: AI Lunch - Noam Brown - January 19, 2016 In-Reply-To: References: Message-ID: Hello everyone, This is a reminder that this lunch and talk is tomorrow. Best, Ellen ---------- Forwarded message ---------- From: Ellen Vitercik Date: Thu, Jan 14, 2016 at 4:48 PM Subject: AI Lunch - Noam Brown - January 19, 2016 To: ai-seminar-announce at cs.cmu.edu Dear faculty and students, We look forward to seeing you this Tuesday, January 19th at noon in *NSH 3305* for AI lunch . There will be Thai food! Noam Brown will speak about Simultaneous Abstraction and Equilibrium Finding. *Abstract: *The leading approach for solving large games of imperfect information is to find a Nash equilibrium in a smaller, abstract version of the game and then map the solution back to the original game. However, it is difficult to know what information and which actions to discard or include in the abstraction without first solving the game, and it is infeasible to solve the game without first abstracting it. We introduce a method that combines abstraction with equilibrium finding by enabling the abstraction to change during run time. This allows an agent to begin learning with a coarse abstraction, and then to strategically refine the abstraction at points that the strategy computed in the current abstraction deems important. The algorithm can switch to the refined abstraction at the cost of a single traversal while provably not having to restart the equilibrium finding. Experiments show it can outperform fixed abstractions at every stage of the run: early on it improves as quickly as equilibrium finding in coarse abstractions, and later it converges to a better solution than does equilibrium finding in fine-grained abstractions. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Jan 20 21:43:55 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 20 Jan 2016 21:43:55 -0500 Subject: [AI Seminar] AI Lunch - Gus Xia - January 26, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, January 26th at noon in NSH 3305 for AI lunch . Gus Xia will give a talk titled "Interactive Artificial Music Performers via Machine Learning." *Abstract:* As both a computer scientist and a musician, I design intelligent systems to understand and extend human musical expression. To understand means to model the musical expression conveyed through acoustic, gestural, and emotional signals. To extend means to use this understanding to create expressive, interactive, and autonomous agents, serving both amateur and professional musicians. In particular, I create interactive artificial performers that are able to perform expressively in concert with humans by learning musicianship from rehearsal experience. This study unifies machine learning and knowledge representation of music structure and performance skills in an HCI framework. In this talk, I will go over the learning techniques and present robot musicians capable of playing collaboratively and reacting to musical nuance with facial and body gestures. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Tue Jan 26 06:37:04 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Tue, 26 Jan 2016 06:37:04 -0500 Subject: [AI Seminar] Fwd: AI Lunch - Gus Xia - January 26, 2016 In-Reply-To: References: Message-ID: Hello everyone, This is a reminder that this lunch and talk is today. Best, Ellen ---------- Forwarded message ---------- From: Ellen Vitercik Date: Wed, Jan 20, 2016 at 9:43 PM Subject: AI Lunch - Gus Xia - January 26, 2016 To: ai-seminar-announce at cs.cmu.edu Dear faculty and students, We look forward to seeing you this Tuesday, January 26th at noon in NSH 3305 for AI lunch . Gus Xia will give a talk titled "Interactive Artificial Music Performers via Machine Learning." *Abstract:* As both a computer scientist and a musician, I design intelligent systems to understand and extend human musical expression. To understand means to model the musical expression conveyed through acoustic, gestural, and emotional signals. To extend means to use this understanding to create expressive, interactive, and autonomous agents, serving both amateur and professional musicians. In particular, I create interactive artificial performers that are able to perform expressively in concert with humans by learning musicianship from rehearsal experience. This study unifies machine learning and knowledge representation of music structure and performance skills in an HCI framework. In this talk, I will go over the learning techniques and present robot musicians capable of playing collaboratively and reacting to musical nuance with facial and body gestures. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Sun Jan 31 23:09:32 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Sun, 31 Jan 2016 23:09:32 -0500 Subject: [AI Seminar] No AI Lunch on February 2, 2016 Message-ID: Dear faculty and students, We will not be having AI lunch this Tuesday, February 2nd. We look forward to seeing you next Tuesday, February 9th for John Dickerson's talk, "Better Matching Markets via Data and Optimization: Evidence from a Nationwide Kidney Exchange." Best, Ellen and Ariel www.cs.cmu.edu/~aiseminar/ -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Thu Feb 4 09:17:32 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Thu, 4 Feb 2016 09:17:32 -0500 Subject: [AI Seminar] AI Lunch - Feb 9th Room Change Message-ID: Dear faculty and students, Please note that AI lunch will be in *NSH 1507* this Tuesday, February 9th at noon. John Dickerson will speak about Better Matching Markets via Data and Optimization: Evidence from a Nationwide Kidney Exchange. *Abstract*: The exchange of indivisible goods without money addresses a variety of constrained economic settings where a medium of exchange?such as money?is considered inappropriate. Participants are either matched directly with another participant or, in more complex domains, in barter cycles and chains with other participants before exchanging their endowed goods. We show that techniques from computer science and operations research, combined with the recent availability of massive data and inexpensive computing, can guide the design of such matching markets and enable the markets by running them in the real world. A key application domain for our work is kidney exchange, an organized market where patients with end-stage renal failure swap willing but incompatible donors. We present new models that address three fundamental dimensions of kidney exchange: (i) uncertainty over the existence of possible trades, (ii) balancing efficiency and fairness, and (iii) inherent dynamism. For each dimension, we design scalable branch-and-price-based integer programming market clearing methods. Next, we combine these dimensions, along with high-level human-provided guidance, into a unified framework for learning to match in a general dynamic setting. This framework, which we coin FutureMatch, takes as input a high-level objective (e.g., "maximize graft survival of transplants over time") decided on by experts, then automatically learns based on data how to make this objective concrete and learns the "means" to accomplish this goal?a task that, in our experience, humans handle poorly. We complement our theoretical models and claims with extensive experiments on real data from the United Network for Organ Sharing (UNOS) Kidney Paired Donation Pilot Program, a large kidney exchange clearinghouse consisting of 60% of the transplant centers in the US. The UNOS exchange uses our algorithms and software to autonomously match donors to patients twice per week. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Feb 8 14:22:42 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 8 Feb 2016 14:22:42 -0500 Subject: [AI Seminar] AI Lunch - Feb 9th Room Change In-Reply-To: References: Message-ID: Hello everyone, This is a reminder that this lunch and talk is tomorrow, Tuesday, February 9th. Best, Ellen On Thu, Feb 4, 2016 at 9:17 AM, Ellen Vitercik wrote: > Dear faculty and students, > > Please note that AI lunch will be in *NSH > 1507* this Tuesday, February 9th at noon. John Dickerson > will speak about Better Matching Markets via > Data and Optimization: Evidence from a Nationwide Kidney Exchange. > > *Abstract*: The exchange of indivisible goods without money addresses a > variety of constrained economic settings where a medium of exchange?such as > money?is considered inappropriate. Participants are either matched directly > with another participant or, in more complex domains, in barter cycles and > chains with other participants before exchanging their endowed goods. We > show that techniques from computer science and operations research, > combined with the recent availability of massive data and inexpensive > computing, can guide the design of such matching markets and enable the > markets by running them in the real world. > > A key application domain for our work is kidney exchange, an organized > market where patients with end-stage renal failure swap willing but > incompatible donors. We present new models that address three fundamental > dimensions of kidney exchange: (i) uncertainty over the existence of > possible trades, (ii) balancing efficiency and fairness, and (iii) inherent > dynamism. For each dimension, we design scalable branch-and-price-based > integer programming market clearing methods. Next, we combine these > dimensions, along with high-level human-provided guidance, into a unified > framework for learning to match in a general dynamic setting. This > framework, which we coin FutureMatch, takes as input a high-level objective > (e.g., "maximize graft survival of transplants over time") decided on by > experts, then automatically learns based on data how to make this objective > concrete and learns the "means" to accomplish this goal?a task that, in our > experience, humans handle poorly. > > We complement our theoretical models and claims with extensive experiments > on real data from the United Network for Organ Sharing (UNOS) Kidney Paired > Donation Pilot Program, a large kidney exchange clearinghouse consisting of > 60% of the transplant centers in the US. The UNOS exchange uses our > algorithms and software to autonomously match donors to patients twice per > week. > > Best, > > Ellen and Ariel > -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Feb 10 19:04:44 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 10 Feb 2016 19:04:44 -0500 Subject: [AI Seminar] AI Lunch - Pengtao Xie - February 16th, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, February 16th, at noon in NSH 3305 for AI lunch . Pengtao Xie will give a talk titled "Diversity-Inducing Learning of Latent Variable Models." *Abstract:* One central task in machine learning (ML) is to extract underlying patterns, structure and knowledge from data. Latent variable models (LVMs) are principled and effective tools to achieve this goal. Due to the dramatic growth of volume and complexity of big data, several new challenges have emerged and cannot be effectively addressed by existing LVMs: (1) How to capture long-tail patterns that carry crucial information when the popularity of patterns is distributed in a power-law fashion? (2) How to reduce model complexity and computational cost without compromising the modeling power of LVMs? (3) How to improve the interpretability and reduce the redundancy of discovered patterns? To addresses the three challenges, we develop a novel regularization technique for LVMs, which controls the geometry of the latent space during learning to enable the learned latent components of LVMs to be diverse, to accomplish long-tail coverage, low redundancy, and better interpretability. In this talk, I will introduce: 1) how the diversity-inducing mutual angular regularizer (MAR) is defined; 2) how to optimize the MAR which is non-convex and non-smooth; 3) a theoretical analysis of why MAR is effective; 4) the applications of MAR in representation learning and distance metric learning. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Feb 15 14:05:41 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 15 Feb 2016 14:05:41 -0500 Subject: [AI Seminar] Fwd: AI Lunch - Pengtao Xie - February 16th, 2016 In-Reply-To: References: Message-ID: Hello everyone, This is a reminder that this lunch and talk is tomorrow, Tuesday, February 16th. Best, Ellen ---------- Forwarded message ---------- From: Ellen Vitercik Date: Wed, Feb 10, 2016 at 7:04 PM Subject: AI Lunch - Pengtao Xie - February 16th, 2016 To: ai-seminar-announce at cs.cmu.edu Dear faculty and students, We look forward to seeing you this Tuesday, February 16th, at noon in NSH 3305 for AI lunch . Pengtao Xie will give a talk titled "Diversity-Inducing Learning of Latent Variable Models." *Abstract:* One central task in machine learning (ML) is to extract underlying patterns, structure and knowledge from data. Latent variable models (LVMs) are principled and effective tools to achieve this goal. Due to the dramatic growth of volume and complexity of big data, several new challenges have emerged and cannot be effectively addressed by existing LVMs: (1) How to capture long-tail patterns that carry crucial information when the popularity of patterns is distributed in a power-law fashion? (2) How to reduce model complexity and computational cost without compromising the modeling power of LVMs? (3) How to improve the interpretability and reduce the redundancy of discovered patterns? To addresses the three challenges, we develop a novel regularization technique for LVMs, which controls the geometry of the latent space during learning to enable the learned latent components of LVMs to be diverse, to accomplish long-tail coverage, low redundancy, and better interpretability. In this talk, I will introduce: 1) how the diversity-inducing mutual angular regularizer (MAR) is defined; 2) how to optimize the MAR which is non-convex and non-smooth; 3) a theoretical analysis of why MAR is effective; 4) the applications of MAR in representation learning and distance metric learning. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Thu Feb 18 23:05:52 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Thu, 18 Feb 2016 23:05:52 -0500 Subject: [AI Seminar] AI Lunch - Professor Scott Fahlman -- February 23rd, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, February 23rd, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . On Tuesday, Professor Scott Fahlman will give a talk titled "Knowledge-Based AI Using Scone." *Abstract*: In the early days of the AI field, the focus was mostly on symbolic knowledge, reasoning, and search, plus a little bit of machine learning to gather the necessary knowledge. More recently, the focus has shifted almost completely to ML, Big Data, and Deep Learning, and there has been some exciting progress in these areas. I will argue that for human-like, human-level AI, we are going to need both approaches: the ML stuff to handle the sensory-motor tasks, and the symbolic stuff for semantics, complex reasoning, and "conscious" thought. In the remainder of the talk, I will give a high-level overview of the open-source Scone knowledge-base system, which my research group has been working on for most of the last decade. I will describe how Scone handles inheritance of information through the "is a" hierarchy, default reasoning with exceptions, and statements about statements, and how we scale the system up to millions of entities and statements -- on a laptop. Perhaps the most unusual feature of Scone is its multiple-context mechanism, which allows us to represent many slightly different world-models within the same knowledge base. Scone's contexts are used for modeling and reasoning about information that changes from one time-point to another; hypotheses and counter-factuals; the different knowledge-states and belief-states of various characters; lies and deception; and in many other ways. Contexts are our secret weapon and our Swiss Army Knife. Finally, I will give a quick overview of the Scone project's current status and what we are working on, now and in the future. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Feb 22 18:01:28 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 22 Feb 2016 18:01:28 -0500 Subject: [AI Seminar] Fwd: AI Lunch - Professor Scott Fahlman -- February 23rd, 2016 In-Reply-To: References: Message-ID: Hello everyone, This is a reminder that this lunch and talk is tomorrow, Tuesday, February 23rd. Best, Ellen ---------- Forwarded message ---------- From: Ellen Vitercik Date: Thu, Feb 18, 2016 at 11:05 PM Subject: AI Lunch - Professor Scott Fahlman -- February 23rd, 2016 To: ai-seminar-announce at cs.cmu.edu Dear faculty and students, We look forward to seeing you this Tuesday, February 23rd, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . On Tuesday, Professor Scott Fahlman will give a talk titled "Knowledge-Based AI Using Scone." *Abstract*: In the early days of the AI field, the focus was mostly on symbolic knowledge, reasoning, and search, plus a little bit of machine learning to gather the necessary knowledge. More recently, the focus has shifted almost completely to ML, Big Data, and Deep Learning, and there has been some exciting progress in these areas. I will argue that for human-like, human-level AI, we are going to need both approaches: the ML stuff to handle the sensory-motor tasks, and the symbolic stuff for semantics, complex reasoning, and "conscious" thought. In the remainder of the talk, I will give a high-level overview of the open-source Scone knowledge-base system, which my research group has been working on for most of the last decade. I will describe how Scone handles inheritance of information through the "is a" hierarchy, default reasoning with exceptions, and statements about statements, and how we scale the system up to millions of entities and statements -- on a laptop. Perhaps the most unusual feature of Scone is its multiple-context mechanism, which allows us to represent many slightly different world-models within the same knowledge base. Scone's contexts are used for modeling and reasoning about information that changes from one time-point to another; hypotheses and counter-factuals; the different knowledge-states and belief-states of various characters; lies and deception; and in many other ways. Contexts are our secret weapon and our Swiss Army Knife. Finally, I will give a quick overview of the Scone project's current status and what we are working on, now and in the future. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Feb 24 13:47:26 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 24 Feb 2016 13:47:26 -0500 Subject: [AI Seminar] AI Seminar - Professor Leslie Pack Kaelbling - March 1, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, March 1st, at noon in NSH 3305 for AI Seminar. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . Leslie Pack Kaelbling , MIT Professor of Computer Science and Engineering, will give a talk titled "Making Robots Behave." *Abstract:* The fields of AI and robotics have made great improvements in many individual subfields, including in motion planning, symbolic planning, probabilistic reasoning, perception, and learning. Our goal is to develop an integrated approach to solving very large problems that are hopelessly intractable to solve optimally. We make a number of approximations during planning, including serializing subtasks, factoring distributions, and determinizing stochastic dynamics, but regain robustness and effectiveness through a continuous state-estimation and replanning process. I will describe our initial approach to this problem, as well as recent work on improving correctness and efficiency through learning. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Feb 29 10:15:20 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 29 Feb 2016 10:15:20 -0500 Subject: [AI Seminar] Fwd: AI Seminar - Professor Leslie Pack Kaelbling - March 1, 2016 In-Reply-To: References: Message-ID: Hello everyone, This is a reminder that this talk is tomorrow, Tuesday, March 1st. Best, Ellen ---------- Forwarded message ---------- From: Ellen Vitercik Date: Wed, Feb 24, 2016 at 1:47 PM Subject: AI Seminar - Professor Leslie Pack Kaelbling - March 1, 2016 To: ai-seminar-announce at cs.cmu.edu Dear faculty and students, We look forward to seeing you this Tuesday, March 1st, at noon in NSH 3305 for AI Seminar. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . Leslie Pack Kaelbling , MIT Professor of Computer Science and Engineering, will give a talk titled "Making Robots Behave." *Abstract:* The fields of AI and robotics have made great improvements in many individual subfields, including in motion planning, symbolic planning, probabilistic reasoning, perception, and learning. Our goal is to develop an integrated approach to solving very large problems that are hopelessly intractable to solve optimally. We make a number of approximations during planning, including serializing subtasks, factoring distributions, and determinizing stochastic dynamics, but regain robustness and effectiveness through a continuous state-estimation and replanning process. I will describe our initial approach to this problem, as well as recent work on improving correctness and efficiency through learning. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Mar 2 22:33:36 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 2 Mar 2016 22:33:36 -0500 Subject: [AI Seminar] AI Lunch -- Nisarg Shah -- March 8th, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, March 8th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . *We are looking for someone to give a talk on March 22nd.* On Tuesday, Nisarg Shah will give a talk titled "Optimal Social Decision Making." *Abstract*: How can computers help ordinary people make collective decisions about real-life dilemmas, like which restaurant to go to with friends, or even how to divide an inheritance? I will present an optimization-driven approach that draws on ideas from AI, theoretical computer science, and economic theory, and illustrate it through my research in computational social choice and computational fair division. In both areas, I will make a special effort to demonstrate how fundamental theoretical questions underlie the design and implementation of deployed services that are already used by tens of thousands of people (spliddit.org), as well as upcoming services (robovote.org). Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From ellen.vitercik at gmail.com Tue Mar 8 07:11:30 2016 From: ellen.vitercik at gmail.com (ellen.vitercik at gmail.com) Date: Tue, 8 Mar 2016 06:11:30 -0600 Subject: [AI Seminar] Fwd: AI Lunch -- Nisarg Shah -- March 8th, 2016 References: Message-ID: <56E6C7AF-FD4A-45DF-A0BE-F02FAFAA39DA@gmail.com> Hello everyone, This is a reminder that this talk is today. Best, Ellen Begin forwarded message: > From: Ellen Vitercik > Date: March 2, 2016 at 9:33:36 PM CST > To: ai-seminar-announce at cs.cmu.edu > Subject: AI Lunch -- Nisarg Shah -- March 8th, 2016 > > Dear faculty and students, > > We look forward to seeing you this Tuesday, March 8th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage. > > On Tuesday, Nisarg Shah will give a talk titled "Optimal Social Decision Making." > > Abstract: How can computers help ordinary people make collective decisions about real-life dilemmas, like which restaurant to go to with friends, or even how to divide an inheritance? I will present an optimization-driven approach that draws on ideas from AI, theoretical computer science, and economic theory, and illustrate it through my research in computational social choice and computational fair division. In both areas, I will make a special effort to demonstrate how fundamental theoretical questions underlie the design and implementation of deployed services that are already used by tens of thousands of people (spliddit.org), as well as upcoming services (robovote.org). > > Best, > Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Thu Mar 10 10:58:11 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Thu, 10 Mar 2016 10:58:11 -0500 Subject: [AI Seminar] No AI Lunch -- March 15, 2016 Message-ID: Dear faculty and students, There will be no AI lunch on March 15th because of the CSD Open House for admitted PhD students. To learn more about the seminar and lunch, please visit the AI Lunch webpage . There are 4 more opportunities to give a talk this semester: *April 5th, April 12th, May 10th, and May 17th. *Please write to me and Ariel (vitercik at cs.cmu.edu, arielpro at cs.cmu.edu) if you are interested in giving a talk. Best, Ellen -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Fri Mar 11 13:51:21 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Fri, 11 Mar 2016 13:51:21 -0500 Subject: [AI Seminar] AI/ML area lunch -- Monday 12-1:30 -- GHC 2109 Message-ID: Dear faculty and PhD students, We hope to see you at the AI/ML area lunch on Monday, March 14th, from 12-1:30 in GHC 2109. This is an important part of the CSD Open House for prospective PhD students. John Dickerson , Phil Thomas and Richard Wang will be giving short talks and there will be time for socializing as well. Please RSVP to Charlotte (cc'ed) if you have not done so already so that there will be lunch for you. Best, Ellen -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Mar 16 13:23:05 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 16 Mar 2016 13:23:05 -0400 Subject: [AI Seminar] AI Lunch -- 3/22 Room Change -- Christian Kroer Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, March 22nd, at noon in *NSH 1507* for AI lunch. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . *We are looking for someone to give a talk on April 5th.* On Tuesday, Christian Kroer will give a talk titled "Arbitrage-free Combinatorial Market Making via Integer Programming." *Abstract*: We present a new combinatorial market maker that operates arbitrage-free combinatorial markets specified by integer programs. The problem of arbitrage-free pricing, while maintaining a bound on the subsidy provided by the market maker, is #P-hard in the worst-case. However, we posit that the typical case might be amenable to modern integer programming (IP) solvers. At the crux of our method is the Frank-Wolfe (conditional gradient) algorithm which is used to implement a Bregman projection aligned with the market maker's cost function, using an IP solver as an oracle. We demonstrate the tractability and improved accuracy of our approach on real-world prediction market data from combinatorial bets placed on the 2010 NCAA Men's Division I Basketball Tournament, where the outcome space is of the size 2^63. To our knowledge, this is the first implementation and empirical evaluation of an arbitrage-free market on this scale. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Mar 21 23:23:50 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 21 Mar 2016 23:23:50 -0400 Subject: [AI Seminar] Fwd: AI Lunch -- 3/22 Room Change -- Christian Kroer In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, March 22nd. Best, Ellen ---------- Forwarded message ---------- From: Ellen Vitercik Date: Wed, Mar 16, 2016 at 1:23 PM Subject: AI Lunch -- 3/22 Room Change -- Christian Kroer To: ai-seminar-announce at cs.cmu.edu Dear faculty and students, We look forward to seeing you this Tuesday, March 22nd, at noon in *NSH 1507* for AI lunch. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . *We are looking for someone to give a talk on April 5th.* On Tuesday, Christian Kroer will give a talk titled "Arbitrage-free Combinatorial Market Making via Integer Programming." *Abstract*: We present a new combinatorial market maker that operates arbitrage-free combinatorial markets specified by integer programs. The problem of arbitrage-free pricing, while maintaining a bound on the subsidy provided by the market maker, is #P-hard in the worst-case. However, we posit that the typical case might be amenable to modern integer programming (IP) solvers. At the crux of our method is the Frank-Wolfe (conditional gradient) algorithm which is used to implement a Bregman projection aligned with the market maker's cost function, using an IP solver as an oracle. We demonstrate the tractability and improved accuracy of our approach on real-world prediction market data from combinatorial bets placed on the 2010 NCAA Men's Division I Basketball Tournament, where the outcome space is of the size 2^63. To our knowledge, this is the first implementation and empirical evaluation of an arbitrage-free market on this scale. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Sat Mar 26 13:41:18 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Sat, 26 Mar 2016 13:41:18 -0400 Subject: [AI Seminar] AI Lunch -- Shayan Doroudi -- March 29, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, March 29th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . *We are looking for someone to give a talk on April 12th.* On Tuesday, Shayan Doroudi will give a talk titled "Importance Sampling for Fair Policy Selection." *Abstract*: Importance sampling is a statistical technique that is used in batch reinforcement learning settings to give unbiased estimates of how well a policy will perform given data from another policy. In addition to evaluating policies, importance sampling has also been used for policy selection and policy search. In this talk, I show that importance sampling is unfair when used to choose policies; that is, in some cases it chooses the worse of two choices more than half of the time. I present several (possibly counterintuitive) examples of where this unfairness may be of practical concern. I then show that, in theory, we can make fair decisions with importance sampling by restricting attention to a particular class of policies. Using insights gathered from the theory, I present a practical policy search algorithm that uses importance sampling with a novel form of regularization. This is joint work with Emma Brunskill and Phil Thomas. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Mar 28 14:42:27 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 28 Mar 2016 14:42:27 -0400 Subject: [AI Seminar] Fwd: AI Lunch -- Shayan Doroudi -- March 29, 2016 In-Reply-To: References: Message-ID: Hello everyone, This is a reminder that this talk is tomorrow. Best, Ellen ---------- Forwarded message ---------- From: Ellen Vitercik Date: Sat, Mar 26, 2016 at 1:41 PM Subject: AI Lunch -- Shayan Doroudi -- March 29, 2016 To: ai-seminar-announce at cs.cmu.edu Dear faculty and students, We look forward to seeing you this Tuesday, March 29th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . *We are looking for someone to give a talk on April 12th.* On Tuesday, Shayan Doroudi will give a talk titled "Importance Sampling for Fair Policy Selection." *Abstract*: Importance sampling is a statistical technique that is used in batch reinforcement learning settings to give unbiased estimates of how well a policy will perform given data from another policy. In addition to evaluating policies, importance sampling has also been used for policy selection and policy search. In this talk, I show that importance sampling is unfair when used to choose policies; that is, in some cases it chooses the worse of two choices more than half of the time. I present several (possibly counterintuitive) examples of where this unfairness may be of practical concern. I then show that, in theory, we can make fair decisions with importance sampling by restricting attention to a particular class of policies. Using insights gathered from the theory, I present a practical policy search algorithm that uses importance sampling with a novel form of regularization. This is joint work with Emma Brunskill and Phil Thomas. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Mar 30 22:10:20 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 30 Mar 2016 22:10:20 -0400 Subject: [AI Seminar] AI Lunch -- Yair Zick -- April 5th, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, April 5th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . *We are looking for someone to give a talk on May 10th.* On Tuesday, Yair Zick will give a talk titled "Towards a Value Theory for Algorithmic Transparency." *Abstract*: Algorithmic systems that employ machine learning play an ever-increasing role in making substantive decisions in modern society, ranging from online personalization to insurance and credit decisions to predictive policing. But their decision making processes are often opaque -- it is difficult to explain why a certain decision was made -- thus raising concerns about inadvertent introduction of harms. We describe a new research agenda, applying game-theoretic centrality to the algorithmic transparency problem. We develop a formal foundation to improve the transparency of such decision-making systems that operate over large volumes of personal information about individuals. First, we describe an axiomatic approach to measuring feature importance in datasets; that is, we derive a function that uniquely satisfies a set of reasonable properties for the measurement of feature influence. Next, we introduce a family of Quantitative Input Influence (QII) measures that capture the degree of influence of inputs on outputs of machine learning algorithms. Our causal QII measures carefully account for correlations among inputs and capture input influence on aggregate effects on groups of individuals (e.g., disparate impact based on race). The QII measures also capture the joint and marginal influence of a set of inputs on outputs using an aggregation method with a strong theoretical justification. Apart from demonstrating general trends in a system, QII guides the construction of personalized transparency reports that provide insights into an individual's classification outcomes. Our empirical validation demonstrates that our QII measures are a useful transparency mechanism when black box access to the learning system is available; in particular, they provide better explanations than standard associative measures for a host of scenarios that we consider. This work is based on two papers Amit Datta, Anupam Datta, Ariel D. Procaccia and Yair Zick "Influence in Classification via Cooperative Game Theory", appeared in the 24th International Joint Conference on Artificial Intelligence (IJCAI 2015). Anupam Datta, Shayak Sen and Yair Zick "Algorithmic Transparency via Quantitative Input Influence: Theory and Experiments with Learning Systems", to appear in the 37th IEEE Symposium on Security and Privacy (Oakland 2016). Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Apr 4 10:12:06 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 4 Apr 2016 10:12:06 -0400 Subject: [AI Seminar] Fwd: AI Lunch -- Yair Zick -- April 5th, 2016 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, April 5th. Best, Ellen ---------- Forwarded message ---------- From: Ellen Vitercik Date: Wed, Mar 30, 2016 at 10:10 PM Subject: AI Lunch -- Yair Zick -- April 5th, 2016 To: ai-seminar-announce at cs.cmu.edu Dear faculty and students, We look forward to seeing you this Tuesday, April 5th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . *We are looking for someone to give a talk on May 10th.* On Tuesday, Yair Zick will give a talk titled "Towards a Value Theory for Algorithmic Transparency." *Abstract*: Algorithmic systems that employ machine learning play an ever-increasing role in making substantive decisions in modern society, ranging from online personalization to insurance and credit decisions to predictive policing. But their decision making processes are often opaque -- it is difficult to explain why a certain decision was made -- thus raising concerns about inadvertent introduction of harms. We describe a new research agenda, applying game-theoretic centrality to the algorithmic transparency problem. We develop a formal foundation to improve the transparency of such decision-making systems that operate over large volumes of personal information about individuals. First, we describe an axiomatic approach to measuring feature importance in datasets; that is, we derive a function that uniquely satisfies a set of reasonable properties for the measurement of feature influence. Next, we introduce a family of Quantitative Input Influence (QII) measures that capture the degree of influence of inputs on outputs of machine learning algorithms. Our causal QII measures carefully account for correlations among inputs and capture input influence on aggregate effects on groups of individuals (e.g., disparate impact based on race). The QII measures also capture the joint and marginal influence of a set of inputs on outputs using an aggregation method with a strong theoretical justification. Apart from demonstrating general trends in a system, QII guides the construction of personalized transparency reports that provide insights into an individual's classification outcomes. Our empirical validation demonstrates that our QII measures are a useful transparency mechanism when black box access to the learning system is available; in particular, they provide better explanations than standard associative measures for a host of scenarios that we consider. This work is based on two papers Amit Datta, Anupam Datta, Ariel D. Procaccia and Yair Zick "Influence in Classification via Cooperative Game Theory", appeared in the 24th International Joint Conference on Artificial Intelligence (IJCAI 2015). Anupam Datta, Shayak Sen and Yair Zick "Algorithmic Transparency via Quantitative Input Influence: Theory and Experiments with Learning Systems", to appear in the 37th IEEE Symposium on Security and Privacy (Oakland 2016). Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Apr 11 07:30:35 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 11 Apr 2016 07:30:35 -0400 Subject: [AI Seminar] AI Lunch -- Shiva Kaul -- April 12, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, April 12th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . *We are looking for someone to give a talk on May 10th*. On Tuesday, Shiva Kaul will give a talk titled "Agnostic linear classification." *Abstract:* We introduce a simple new algorithm for improper, agnostic learning of halfspaces, a problem closely related to learning sparse parities with noise. It uses exponentially less data than previous algorithms, particularly ones based on fitting polynomials. It is provably correct in a wide range of settings, but not necessarily fast. The algorithm is very practical and achieves good experimental performance for both natural and artificial problems. The lynchpin of this algorithm is a new hypothesis class called smooth lists of halfspaces. They are more flexible than halfspaces, but do not require more data to train in the worst case. These new classifiers enable an iterative approach which is fundamentally different than update rules such as perceptron and multiplicative weights. Our analysis involves a dual interpretation of the algorithm as a dynamical system. Joint work with Geoff Gordon. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From arielpro at gmail.com Mon Apr 11 11:03:28 2016 From: arielpro at gmail.com (Ariel Procaccia) Date: Mon, 11 Apr 2016 11:03:28 -0400 Subject: [AI Seminar] Meetings with Rina Dechter on April 26 Message-ID: Hi everyone, Rina Dechter of UCI will give the AI seminar talk on April 26 (see http://www.cs.cmu.edu/~aiseminar/). She will be available for meetings on the day of the talk. If you'd like to meet with her, please email me your constraints. Thanks, Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Apr 13 21:28:44 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 13 Apr 2016 21:28:44 -0400 Subject: [AI Seminar] AI Lunch -- Christopher MacLellan -- April 19, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, April 19th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . *We are looking for someone to give a talk on May 10th*. On Tuesday, Christopher MacLellan will give a talk titled "Using the Apprentice Learner Architecture to model human learning from demonstrations and feedback in digital environments." *Abstract:* Understanding the nature of human intelligence and developing intelligent agents capable of modeling humans are fundamental goals of artificial intelligence research. Prior work modeling human problem solving has explored how hand-constructed domain models (e.g., production-rule models) can be used to explain human behavior. Typically, these models account for how humans improve their problem-solving performance given practice (i.e., speed-up learning), but they do not account for how humans acquire initial domain models. One approach to acquiring domain knowledge that has been explored in the machine learning literature is *Apprentice Learning* (also called programming by demonstration, learning by watching, or imitation learning). Previous work has explored how apprentice learning can be used as an efficient, user-friendly approach to programming computers and robots by providing them with demonstrations rather than computer code. In the current work, we investigate the possibility that computational approaches for apprentice learning can be used to model human learning in digital learning environments, such as intelligent tutoring systems. Towards this end, I present the Apprentice Learner Architecture, which provides a framework for building models of apprentice learning in these digital environments. Next, I show how this architecture can be used to construct models of human learning capable of simulating and predicting human behavior in intelligent tutors. In particular, I show how apprentice learner models can be used to simulate human behavior in a fraction arithmetic tutor and how these simulations can be used to accurately predict the main experimental effects of a human study that used the fractions tutor. Finally, I conclude with directions for future work. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Apr 18 09:32:46 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 18 Apr 2016 09:32:46 -0400 Subject: [AI Seminar] Fwd: AI Lunch -- Christopher MacLellan -- April 19, 2016 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, April 19th. Best, Ellen ---------- Forwarded message ---------- From: Ellen Vitercik Date: Wed, Apr 13, 2016 at 9:28 PM Subject: AI Lunch -- Christopher MacLellan -- April 19, 2016 To: ai-seminar-announce at cs.cmu.edu Dear faculty and students, We look forward to seeing you this Tuesday, April 19th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . *We are looking for someone to give a talk on May 10th*. On Tuesday, Christopher MacLellan will give a talk titled "Using the Apprentice Learner Architecture to model human learning from demonstrations and feedback in digital environments." *Abstract:* Understanding the nature of human intelligence and developing intelligent agents capable of modeling humans are fundamental goals of artificial intelligence research. Prior work modeling human problem solving has explored how hand-constructed domain models (e.g., production-rule models) can be used to explain human behavior. Typically, these models account for how humans improve their problem-solving performance given practice (i.e., speed-up learning), but they do not account for how humans acquire initial domain models. One approach to acquiring domain knowledge that has been explored in the machine learning literature is *Apprentice Learning* (also called programming by demonstration, learning by watching, or imitation learning). Previous work has explored how apprentice learning can be used as an efficient, user-friendly approach to programming computers and robots by providing them with demonstrations rather than computer code. In the current work, we investigate the possibility that computational approaches for apprentice learning can be used to model human learning in digital learning environments, such as intelligent tutoring systems. Towards this end, I present the Apprentice Learner Architecture, which provides a framework for building models of apprentice learning in these digital environments. Next, I show how this architecture can be used to construct models of human learning capable of simulating and predicting human behavior in intelligent tutors. In particular, I show how apprentice learner models can be used to simulate human behavior in a fraction arithmetic tutor and how these simulations can be used to accurately predict the main experimental effects of a human study that used the fractions tutor. Finally, I conclude with directions for future work. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Apr 20 13:44:12 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 20 Apr 2016 13:44:12 -0400 Subject: [AI Seminar] AI Seminar -- Professor Rina Dechter -- April 26, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, April 26th, at noon in NSH 3305 for AI Seminar. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . *We are looking for someone to give a talk on May 10th*. UC Irvine Professor Rina Dechter will give a talk titled "Modern Exact and Approximate Combinatorial Optimization Algorithms (max-product and max-sum-product) for Graphical models." *Abstract: *In this talk I will present several principles behind state of the art algorithms for solving combinatorial optimization tasks defined over graphical models (Bayesian networks, Markov networks, constraint networks, Influence diagrams) and demonstrate their performance on some benchmarks. Specifically I will present branch and bound search algorithms which explore the AND/OR search space over graphical models and thus exploit problem's decomposition (using AND nodes), equivalence (by caching) and pruning irrelevant subspaces via the power of bounding heuristics. In particular I will show how the two ideas of mini-bucket partitioning which relaxes the input problem using node duplication only, combined with linear programming relaxations ideas which optimize cost-shifting/re-parameterization schemes, can yield tight bounding heuristic information within systematic, anytime, search. I will then show our recent extension to solving the far more challenging task of Marginal Map (as time permits). Notably, a solver for finding the most probable explanation (MPE or MAP), embedding these principles has won first place in all time categories in the 2012 PASCAL2 approximate inference challenge, and first or second place in the UAI-2014 competitions. Parts of this work were done jointly with: Radu Marinescu, Kalev Kask, Alex Ihler, Robesrt Mateescu and Lars Otten. *Bio:* Rina Dechter is a professor of Computer Science at the University of California, Irvine. She received her PhD in Computer Science at UCLA in 1985, an MS degree in Applied Mathematics from the Weizmann Institute and a B.S in Mathematics and Statistics from the Hebrew University, Jerusalem. Her research centers on computational aspects of automated reasoning and knowledge representation including search, constraint processing and probabilistic reasoning. Professor Dechter is an author of 'Constraint Processing' published by Morgan Kaufmann, 2003, and 'Reasoning with Probabilistic and Deterministic Graphical Models: Exact Algorithms' by Morgan and Claypool publishers, 2013, has co-authored over 150 research papers, and has served on the editorial boards of: Artificial Intelligence, the Constraint Journal, Journal of Artificial Intelligence Research, Logical Methods in Computer Science (LMCS) and journal of Machine Learning (JLMR). She was awarded the Presidential Young investigator award in 1991, is a fellow of the American association of Artificial Intelligence since 1994, was a Radcliffe Fellowship 2005-2006, received the 2007 Association of Constraint Programming (ACP) research excellence award and is a 2013 Fellow of the ACM. She has been Co-Editor-in-Chief of Artificial Intelligence, since 2011. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From arielpro at cs.cmu.edu Sat Apr 23 16:22:11 2016 From: arielpro at cs.cmu.edu (Ariel Procaccia) Date: Sat, 23 Apr 2016 16:22:11 -0400 Subject: [AI Seminar] Meetings with Robin Hanson on May 3 Message-ID: Hi everyone, Robin Hanson will give the AI seminar talk on May 3 (see http://www.cs.cmu.edu/~aiseminar/ for more details). He will be available for meetings on the day of the talk. If you'd like to meet with him, please email me your constraints. Thanks, Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From nmramesh at cs.cmu.edu Sat Apr 23 16:28:10 2016 From: nmramesh at cs.cmu.edu (Ramesh Nallapati) Date: Sat, 23 Apr 2016 20:28:10 +0000 Subject: [AI Seminar] Unsubscribe In-Reply-To: References: Message-ID: On Sat, Apr 23, 2016, 4:27 PM Ariel Procaccia wrote: > Hi everyone, > > Robin Hanson will give the AI seminar talk on May 3 (see > http://www.cs.cmu.edu/~aiseminar/ for more details). He will be available > for meetings on the day of the talk. If you'd like to meet with him, please > email me your constraints. > > Thanks, > Ariel > -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Apr 25 09:50:55 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 25 Apr 2016 09:50:55 -0400 Subject: [AI Seminar] AI Seminar -- Professor Rina Dechter -- April 26, 2016 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, April 26th. Best, Ellen On Wed, Apr 20, 2016 at 1:44 PM, Ellen Vitercik wrote: > Dear faculty and students, > > We look forward to seeing you this Tuesday, April 26th, at noon in NSH > 3305 for AI Seminar. To learn more about the seminar and lunch, or to > volunteer to give a talk, please visit the AI Lunch webpage > . *We are looking for someone to give > a talk on May 10th*. > > UC Irvine Professor Rina Dechter will > give a talk titled "Modern Exact and Approximate Combinatorial > Optimization Algorithms (max-product and max-sum-product) for Graphical > models." > > *Abstract: *In this talk I will present several principles behind state > of the art algorithms for solving combinatorial optimization tasks defined > over graphical models (Bayesian networks, Markov networks, constraint > networks, Influence diagrams) and demonstrate their performance on some > benchmarks. > > Specifically I will present branch and bound search algorithms which > explore the AND/OR search space over graphical models and thus exploit > problem's decomposition (using AND nodes), equivalence (by caching) and > pruning irrelevant subspaces via the power of bounding heuristics. In > particular I will show how the two ideas of mini-bucket partitioning which > relaxes the input problem using node duplication only, combined with linear > programming relaxations ideas which optimize > cost-shifting/re-parameterization schemes, can yield tight bounding > heuristic information within systematic, anytime, search. I will then show > our recent extension to solving the far more challenging task of Marginal > Map (as time permits). > > Notably, a solver for finding the most probable explanation (MPE or MAP), > embedding these principles has won first place in all time categories in > the 2012 PASCAL2 approximate inference challenge, and first or second place > in the UAI-2014 competitions. > > Parts of this work were done jointly with: Radu Marinescu, Kalev Kask, > Alex Ihler, Robesrt Mateescu and Lars Otten. > > *Bio:* Rina Dechter is a professor of Computer Science at the University > of California, Irvine. She received her PhD in Computer Science at UCLA in > 1985, an MS degree in Applied Mathematics from the Weizmann Institute and a > B.S in Mathematics and Statistics from the Hebrew University, Jerusalem. > Her research centers on computational aspects of automated reasoning and > knowledge representation including search, constraint processing and > probabilistic reasoning. > > Professor Dechter is an author of 'Constraint Processing' published by > Morgan Kaufmann, 2003, and 'Reasoning with Probabilistic and Deterministic > Graphical Models: Exact Algorithms' by Morgan and Claypool publishers, > 2013, has co-authored over 150 research papers, and has served on the > editorial boards of: Artificial Intelligence, the Constraint Journal, > Journal of Artificial Intelligence Research, Logical Methods in Computer > Science (LMCS) and journal of Machine Learning (JLMR). She was awarded the > Presidential Young investigator award in 1991, is a fellow of the American > association of Artificial Intelligence since 1994, was a Radcliffe > Fellowship 2005-2006, received the 2007 Association of Constraint > Programming (ACP) research excellence award and is a 2013 Fellow of the > ACM. She has been Co-Editor-in-Chief of Artificial Intelligence, since 2011. > > Best, > > Ellen and Ariel > -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Apr 27 18:40:43 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 27 Apr 2016 18:40:43 -0400 Subject: [AI Seminar] AI Seminar -- Professor Robin Hanson -- May 3, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, May 3rd, at noon in NSH 3305 for AI Seminar. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . *We are looking for someone to give a talk on May 10th*. George Mason University Professor Robin Hanson will give a talk titled "The Age of Em: Work, Love, and Life when Robots Rule the Earth." *Abstract:* The three most disruptive transitions in history were the introduction of humans, farming, and industry. If a similar transition lies ahead, a good guess for its source is artificial intelligence in the form of whole brain emulations, or "ems," sometime in the next century. Drawing on academic consensus in many disciplines, I outline a baseline scenario set modestly far into a post-em-transition world. I consider computer architecture, energy use, cooling infrastructure, mind speeds, body sizes, security strategies, virtual reality conventions, labor market organization, management focus, job training, career paths, wage competition, identity, retirement, life cycles, reproduction, mating, conversation habits, wealth inequality, city sizes, growth rates, coalition politics, governance, law, and war. To learn more, see Professor Hanson's forthcoming book The Age of Em . *Bio:* Robin Hanson is Assoc. Prof. of economics at George Mason University, and research associate at Future of Humanity Institute of Oxford University. He has a Caltech social science Ph.D., University of Chicago physics M.S and philosophy M.A., 9 years experience as A.I. research programmer, at Lockheed and NASA, 3050 citations, h-index of 25, 60 academic publications, 400 media mentions, 200 invited talks, and 8 million visits to his blog OvercomingBias.com. Oxford Press publishes his The Age of Em: Work, Love and Life When Robots Rule the Earth in June 2016, and The Elephant in the Brain: Hidden Motives in Everyday Life, Kevin Simler co-author, in spring 2017. A prediction market pioneer since 1988, he was architect of first internal corporate markets, at Xanadu in 1990, of the Foresight Exchange since 1994, of DARPA's Policy Analysis Market, from 2001 to 2003, and of IARPA's DAGGRE and SCICAST from 2010 to 2015. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Fri Apr 29 08:44:34 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Fri, 29 Apr 2016 08:44:34 -0400 Subject: [AI Seminar] ROOM CHANGE -- AI Seminar -- Professor Robin Hanson -- May 3, 2016 Message-ID: ROOM CHANGE: Professor Robin Hanson's talk will be in GHC 6115. On Tuesday, May 3rd, at noon, George Mason University Professor Robin Hanson will give a talk titled "The Age of Em: Work, Love, and Life when Robots Rule the Earth." *Abstract:* The three most disruptive transitions in history were the introduction of humans, farming, and industry. If a similar transition lies ahead, a good guess for its source is artificial intelligence in the form of whole brain emulations, or "ems," sometime in the next century. Drawing on academic consensus in many disciplines, I outline a baseline scenario set modestly far into a post-em-transition world. I consider computer architecture, energy use, cooling infrastructure, mind speeds, body sizes, security strategies, virtual reality conventions, labor market organization, management focus, job training, career paths, wage competition, identity, retirement, life cycles, reproduction, mating, conversation habits, wealth inequality, city sizes, growth rates, coalition politics, governance, law, and war. To learn more, see Professor Hanson's forthcoming book The Age of Em . *Bio:* Robin Hanson is Assoc. Prof. of economics at George Mason University, and research associate at Future of Humanity Institute of Oxford University. He has a Caltech social science Ph.D., University of Chicago physics M.S and philosophy M.A., 9 years experience as A.I. research programmer, at Lockheed and NASA, 3050 citations, h-index of 25, 60 academic publications, 400 media mentions, 200 invited talks, and 8 million visits to his blog OvercomingBias.com . Oxford Press publishes his The Age of Em: Work, Love and Life When Robots Rule the Earth in June 2016, and The Elephant in the Brain: Hidden Motives in Everyday Life, Kevin Simler co-author, in spring 2017. A prediction market pioneer since 1988, he was architect of first internal corporate markets, at Xanadu in 1990, of the Foresight Exchange since 1994, of DARPA's Policy Analysis Market, from 2001 to 2003, and of IARPA's DAGGRE and SCICAST from 2010 to 2015. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From ellen.vitercik at gmail.com Tue May 3 10:56:07 2016 From: ellen.vitercik at gmail.com (ellen.vitercik at gmail.com) Date: Tue, 3 May 2016 10:56:07 -0400 Subject: [AI Seminar] Fwd: ROOM CHANGE -- AI Seminar -- Professor Robin Hanson -- May 3, 2016 References: Message-ID: Hello everyone, This is a reminder that this talk is today and the location is GHC 6115. Best, Ellen Begin forwarded message: > From: Ellen Vitercik > Date: April 29, 2016 at 8:44:34 AM EDT > To: ai-seminar-announce at cs.cmu.edu > Subject: ROOM CHANGE -- AI Seminar -- Professor Robin Hanson -- May 3, 2016 > > ROOM CHANGE: > Professor Robin Hanson's talk will be in GHC 6115. > > On Tuesday, May 3rd, at noon, George Mason University Professor Robin Hanson will give a talk titled "The Age of Em: Work, Love, and Life when Robots Rule the Earth." > > Abstract: The three most disruptive transitions in history were the introduction of humans, farming, and industry. If a similar transition lies ahead, a good guess for its source is artificial intelligence in the form of whole brain emulations, or "ems," sometime in the next century. Drawing on academic consensus in many disciplines, I outline a baseline scenario set modestly far into a post-em-transition world. I consider computer architecture, energy use, cooling infrastructure, mind speeds, body sizes, security strategies, virtual reality conventions, labor market organization, management focus, job training, career paths, wage competition, identity, retirement, life cycles, reproduction, mating, conversation habits, wealth inequality, city sizes, growth rates, coalition politics, governance, law, and war. > > To learn more, see Professor Hanson's forthcoming book The Age of Em. > > Bio: Robin Hanson is Assoc. Prof. of economics at George Mason University, and research associate at Future of Humanity Institute of Oxford University. He has a Caltech social science Ph.D., University of Chicago physics M.S and philosophy M.A., 9 years experience as A.I. research programmer, at Lockheed and NASA, 3050 citations, h-index of 25, 60 academic publications, 400 media mentions, 200 invited talks, and 8 million visits to his blog OvercomingBias.com. > > Oxford Press publishes his The Age of Em: Work, Love and Life When Robots Rule the Earth in June 2016, and The Elephant in the Brain: Hidden Motives in Everyday Life, Kevin Simler co-author, in spring 2017. A prediction market pioneer since 1988, he was architect of first internal corporate markets, at Xanadu in 1990, of the Foresight Exchange since 1994, of DARPA's Policy Analysis Market, from 2001 to 2003, and of IARPA's DAGGRE and SCICAST from 2010 to 2015. > > Best, > > Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Thu May 5 19:29:23 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Thu, 5 May 2016 19:29:23 -0400 Subject: [AI Seminar] AI Lunch -- Rogelio Cardona-Rivera -- May 10th, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, May 10th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Rogelio Cardona-Rivera will give a talk titled "Toward the Holodeck: Computational Models of Interactive Narrative and their relation to Human Cognition." *Abstract:* Interactive narratives are used for an ever-expanding array of purposes: educational settings, training simulations, and even organizational behaviors have had narratives woven around them because these are made more compelling in a dramatic framing. Despite their ubiquity, they remain difficult to engineer. One reason is because we lack a precise understanding of human narrative intelligence, which would explain how we interact with stories. In this talk, I will present my approach to developing a computational-cognitive model of narrative affordances, which centers on predicting how users imagine themselves taking actions in an unfolding narrative virtual environment. I will discuss this approach in the context of applications of automated planning and activity recognition to model a user's search of an author's intended meaning. Concluding the talk I will discuss the potential for this approach to enable more engaging narrative experiences, through the next generation of intelligent and adaptive virtual environments. *Speaker Bio:* Rogelio Cardona-Rivera is an ABD Ph.D. Candidate in Computer Science at North Carolina State University. He researches technologies to improve interactive narrative design and development through artificial intelligence and cognitive science. Rogelio has been a recipient of the National GEM Fellowship, and the Department of Energy's Computational Science Graduate Fellowship. Rogelio has published at diverse, high-impact venues in and around intelligent narrative technologies. His work has been recognized with a Best Paper Award at the International Conference on Interactive Digital Storytelling (ICIDS) in 2012, a Best Student Paper on a Cognitive Science Topic at the Workshop on Computational Models of Narrative in 2012, and an Honorable Mention for Best Paper at the Computer-Human Interaction Conference in 2016. He has served on numerous program committees, and will co-chair the Intelligent Narrative Technologies track at ICIDS in 2016. Rogelio received his B.Sc. in Computer Engineering from the University of Puerto Rico at Mayag?ez and has interned as a computational narratologist at Sandia National Laboratories and Disney Research. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon May 9 17:38:49 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 9 May 2016 17:38:49 -0400 Subject: [AI Seminar] AI Lunch -- Rogelio Cardona-Rivera -- May 10th, 2016 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, May 10th. Best, Ellen On Thu, May 5, 2016 at 7:29 PM, Ellen Vitercik wrote: > Dear faculty and students, > > We look forward to seeing you this Tuesday, May 10th, at noon in NSH 3305 > for AI lunch. To learn more about the seminar and lunch, please visit the AI > Lunch webpage . > > On Tuesday, Rogelio Cardona-Rivera will give a talk titled "Toward the > Holodeck: Computational Models of Interactive Narrative and their relation > to Human Cognition." > > *Abstract:* Interactive narratives are used for an ever-expanding array > of purposes: educational settings, training simulations, and even > organizational behaviors have had narratives woven around them because > these are made more compelling in a dramatic framing. Despite their > ubiquity, they remain difficult to engineer. One reason is because we lack > a precise understanding of human narrative intelligence, which would > explain how we interact with stories. In this talk, I will present my > approach to developing a computational-cognitive model of narrative > affordances, which centers on predicting how users imagine themselves > taking actions in an unfolding narrative virtual environment. I will > discuss this approach in the context of applications of automated planning > and activity recognition to model a user's search of an author's intended > meaning. Concluding the talk I will discuss the potential for this approach > to enable more engaging narrative experiences, through the next generation > of intelligent and adaptive virtual environments. > > *Speaker Bio:* Rogelio Cardona-Rivera is an ABD Ph.D. Candidate in > Computer Science at North Carolina State University. He researches > technologies to improve interactive narrative design and development > through artificial intelligence and cognitive science. Rogelio has been a > recipient of the National GEM Fellowship, and the Department of Energy's > Computational Science Graduate Fellowship. Rogelio has published at > diverse, high-impact venues in and around intelligent narrative > technologies. His work has been recognized with a Best Paper Award at the > International Conference on Interactive Digital Storytelling (ICIDS) in > 2012, a Best Student Paper on a Cognitive Science Topic at the Workshop on > Computational Models of Narrative in 2012, and an Honorable Mention for > Best Paper at the Computer-Human Interaction Conference in 2016. He has > served on numerous program committees, and will co-chair the Intelligent > Narrative Technologies track at ICIDS in 2016. Rogelio received his B.Sc. > in Computer Engineering from the University of Puerto Rico at Mayag?ez and > has interned as a computational narratologist at Sandia National > Laboratories and Disney Research. > > Best, > Ellen and Ariel > -------------- next part -------------- An HTML attachment was scrubbed... URL: From mmv at cs.cmu.edu Thu Jun 23 14:13:46 2016 From: mmv at cs.cmu.edu (Manuela Veloso) Date: Thu, 23 Jun 2016 19:13:46 +0100 Subject: [AI Seminar] Fwd: IBM Watson A.I. XPRIZE Opens for Registration References: Message-ID: <945015D9-D7BA-43CF-8469-83113E97BD36@cs.cmu.edu> Sent from my iPhone Begin forwarded message: > From: XPRIZE > Date: June 23, 2016 at 6:27:10 PM GMT+1 > To: Manuela Veloso > Subject: IBM Watson A.I. XPRIZE Opens for Registration > Reply-To: XPRIZE > > > View this email in your browser > > > > IBM WATSON AI XPRIZE AIMING TO ACCELERATE THE DEVELOPMENT AND ADOPTION OF ARTIFICIAL INTELLIGENCE TO TACKLE THE WORLD?S GRAND CHALLENGES > XPRIZE is pleased to announce the registration for the IBM Watson AI XPRIZE, a global competition demonstrating how humans can collaborate with powerful cognitive technologies to tackle some of the world?s grand challenges. > > If you are interested in creating or joining a team, please visit our competition portal. > > If you would like to help in ways other than competing for the prize, please click here. > > Learn More >> > > > JOIN THE CONVERSATION > > Copyright ? 2016 XPRIZE, All rights reserved. > You are receiving this email because you opted in at our website, http://www.xprize.org > > Our mailing address is: > XPRIZE > 800 Corporate Pointe, Suite 350 > Culver City, California 90230 > > Add us to your address book > > > unsubscribe from this list update subscription preferences > -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Sat Jul 23 17:12:36 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Sat, 23 Jul 2016 17:12:36 -0400 Subject: [AI Seminar] AI Lunch Fall Semester Talk Signup Message-ID: Dear faculty and students, We are beginning to plan the AI lunch fall semester schedule, and we would appreciate all volunteers who are interested in giving a talk. Lunch will be every Tuesday at noon beginning on August 30th, and talks should be about 45-50 minutes long. Please email me (vitercik at cs.cmu.edu) and Ariel (arielpro at cs.cmu.edu) to sign up. Thank you! Ellen Vitercik -------------- next part -------------- An HTML attachment was scrubbed... URL: From arielpro at cs.cmu.edu Sun Aug 14 17:48:45 2016 From: arielpro at cs.cmu.edu (Ariel Procaccia) Date: Mon, 15 Aug 2016 00:48:45 +0300 Subject: [AI Seminar] AI Lunch Fall Semester Talk Signup In-Reply-To: References: Message-ID: Dear all, We've had only one person actively volunteer to give a talk so far. The continued success of the seminar depends on you! If you're a student or postdoc and you haven't given a talk last semester (or you have but there's really exciting new stuff you want to talk about), please sign up by emailing Ellen and me. If you're a faculty member, please encourage your students and postdocs to sign up, or even sign up to give a talk yourself if you wish. By the way, as in the spring, the seminar will also feature several talks by prominent external speakers. Stay tuned... Cheers, Ariel On Sun, Jul 24, 2016 at 12:12 AM, Ellen Vitercik wrote: > Dear faculty and students, > > We are beginning to plan the AI lunch fall semester schedule, and we would > appreciate all volunteers who are interested in giving a talk. Lunch will > be every Tuesday at noon beginning on August 30th, and talks should be > about 45-50 minutes long. > > Please email me (vitercik at cs.cmu.edu) and Ariel (arielpro at cs.cmu.edu) to > sign up. > > Thank you! > > Ellen Vitercik > -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Sat Aug 20 09:43:07 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Sat, 20 Aug 2016 09:43:07 -0400 Subject: [AI Seminar] AI Lunch -- Andrew Moore -- August 30th, 2016 Message-ID: Dear faculty and students, We look forward to seeing you on Tuesday, August 30th for the first AI lunch of the semester! Lunch will be at noon in *NSH 1507*. (On subsequent Tuesdays, we will meet in NSH 3305, unless otherwise noted.) Please visit the AI Lunch webpage to see the lineup of talks we have scheduled for the next few months. If you are interested in giving a talk, feel free to email me and Ariel (arielpro at cs.cmu.edu). On August 30th, we are excited to have Andrew Moore , Dean of the School of Computer Science, present this semester's first talk. His talk is titled "Explaining AI at CMU to the rest of the world." *Abstract:* Within CMU and SCS we are very confident that we are helping lead the world in AI and other forms of advanced computer science / statistics / optimization / game theory to make the world a much better, funner and safer place. This is a direct legacy from two of the four founders of AI: Newell and Simon. How do we explain what's going on right now in a way that is interesting and compelling? How do we explain it in different contexts? With a lot of help from Manuela, Martial, Jaime and others I have a story assembled which I've been using to try to spread the word. I need feedback on it and ideas for which audiences need it tweaked in which direction. I would like to spend 30 mins giving the pitch and then spend 30 mins discussing how to improve it, and also, of course, whether it might have useful content for the other CMU-AI folks to use when needed. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Tue Aug 23 21:48:51 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Tue, 23 Aug 2016 21:48:51 -0400 Subject: [AI Seminar] AI Lunch -- Room Change -- August 30 Message-ID: Dear faculty and students, On August 30th, AI lunch will be held at noon in *GHC 6115*, not NSH 1507, as previously planned. As I described in the previous announcement, Andrew Moore will present a talk titled "Explaining AI at CMU to the rest of the world." *Abstract: *Within CMU and SCS we are very confident that we are helping lead the world in AI and other forms of advanced computer science / statistics / optimization / game theory to make the world a much better, funner and safer place. This is a direct legacy from two of the four founders of AI: Newell and Simon. How do we explain what's going on right now in a way that is interesting and compelling? How do we explain it in different contexts? With a lot of help from Manuela, Martial, Jaime and others I have a story assembled which I've been using to try to spread the word. I need feedback on it and ideas for which audiences need it tweaked in which direction. I would like to spend 30 mins giving the pitch and then spend 30 mins discussing how to improve it, and also, of course, whether it might have useful content for the other CMU-AI folks to use when needed. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From arielpro at cs.cmu.edu Fri Aug 26 10:46:08 2016 From: arielpro at cs.cmu.edu (Ariel Procaccia) Date: Fri, 26 Aug 2016 10:46:08 -0400 Subject: [AI Seminar] Fwd: PhD Linear Programming course at Tepper In-Reply-To: <809f4414-055d-d065-bdc4-d6d204855aa1@andrew.cmu.edu> References: <809f4414-055d-d065-bdc4-d6d204855aa1@andrew.cmu.edu> Message-ID: These courses might be of interest to many of you (or your students). Best, Ariel ---------- Forwarded message ---------- From: Fatma Kilinc-Karzan Date: Thu, Aug 25, 2016 at 12:47 PM Subject: PhD Linear Programming course at Tepper To: Ariel Procaccia Hi Ariel, I hope your summer went well and you are off to a great start of the semester. I will be teaching the PhD Linear Programming course this Fall at Tepper. I think some of the students at CS might be interested in this course. I will very much appreciate if you can disseminate this information. In Fall 2016 (mini-1), I will be teaching a PhD level Linear Programming course (47-834). This course is a required course for PhD students in our Operations Research, Algorithms, Combinatorics and Optimization and Operations Management PhD programs. In addition to these students, we usually have various students from Computer Science, Robotics, Chemical Engineering and other engineering and business areas. This course serves the basis for the rest of the PhD level Optimization courses we teach in these PhD programs and at Tepper. The course syllabus is attached to this email, interested students are welcome to get in touch with me. Thanks, Fatma -- Fatma Kilinc Karzan Associate Professor of Operations Research Tepper School of Business Carnegie Mellon University fkilinc at andrew.cmu.edu (+1) 412 268 9198 -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: 47834 Syllabus (Fall 2016).pdf Type: application/pdf Size: 95429 bytes Desc: not available URL: From vitercik at cs.cmu.edu Mon Aug 29 08:39:39 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 29 Aug 2016 08:39:39 -0400 Subject: [AI Seminar] Fwd: AI Lunch -- Room Change -- August 30 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, August 30th. ---------- Forwarded message ---------- From: Ellen Vitercik Date: Tue, Aug 23, 2016 at 9:48 PM Subject: AI Lunch -- Room Change -- August 30 To: ai-seminar-announce at cs.cmu.edu Dear faculty and students, On August 30th, AI lunch will be held at noon in *GHC 6115*, not NSH 1507, as previously planned. As I described in the previous announcement, Andrew Moore will present a talk titled "Explaining AI at CMU to the rest of the world." *Abstract: *Within CMU and SCS we are very confident that we are helping lead the world in AI and other forms of advanced computer science / statistics / optimization / game theory to make the world a much better, funner and safer place. This is a direct legacy from two of the four founders of AI: Newell and Simon. How do we explain what's going on right now in a way that is interesting and compelling? How do we explain it in different contexts? With a lot of help from Manuela, Martial, Jaime and others I have a story assembled which I've been using to try to spread the word. I need feedback on it and ideas for which audiences need it tweaked in which direction. I would like to spend 30 mins giving the pitch and then spend 30 mins discussing how to improve it, and also, of course, whether it might have useful content for the other CMU-AI folks to use when needed. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From ravi at andrew.cmu.edu Mon Aug 29 18:34:17 2016 From: ravi at andrew.cmu.edu (R Ravi) Date: Mon, 29 Aug 2016 22:34:17 +0000 Subject: [AI Seminar] PhD Linear Programming course at Tepper In-Reply-To: References: <809f4414-055d-d065-bdc4-d6d204855aa1@andrew.cmu.edu> Message-ID: <5e8b8a3864b1476093e88a344252334e@PGH-MSGMLT-02.andrew.ad.cmu.edu> In the same vein, I start teaching an introductory graph theory class at Tepper (syllabus enclosed) which begins tomorrow. All interested students are welcome. Ravi From: Theory-announce [mailto:theory-announce-bounces at mailman.srv.cs.cmu.edu] On Behalf Of Ariel Procaccia Sent: Friday, August 26, 2016 10:46 AM To: theory-announce at cs.cmu.edu; ai-seminar-announce at cs.cmu.edu Subject: Fwd: PhD Linear Programming course at Tepper These courses might be of interest to many of you (or your students). Best, Ariel ---------- Forwarded message ---------- From: Fatma Kilinc-Karzan > Date: Thu, Aug 25, 2016 at 12:47 PM Subject: PhD Linear Programming course at Tepper To: Ariel Procaccia > Hi Ariel, I hope your summer went well and you are off to a great start of the semester. I will be teaching the PhD Linear Programming course this Fall at Tepper. I think some of the students at CS might be interested in this course. I will very much appreciate if you can disseminate this information. In Fall 2016 (mini-1), I will be teaching a PhD level Linear Programming course (47-834). This course is a required course for PhD students in our Operations Research, Algorithms, Combinatorics and Optimization and Operations Management PhD programs. In addition to these students, we usually have various students from Computer Science, Robotics, Chemical Engineering and other engineering and business areas. This course serves the basis for the rest of the PhD level Optimization courses we teach in these PhD programs and at Tepper. The course syllabus is attached to this email, interested students are welcome to get in touch with me. Thanks, Fatma -- Fatma Kilinc Karzan Associate Professor of Operations Research Tepper School of Business Carnegie Mellon University fkilinc at andrew.cmu.edu (+1) 412 268 9198 -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: outline-GT2016.pdf Type: application/pdf Size: 67284 bytes Desc: outline-GT2016.pdf URL: From haeupler at cs.cmu.edu Tue Aug 30 12:11:25 2016 From: haeupler at cs.cmu.edu (Bernhard Haeupler) Date: Tue, 30 Aug 2016 12:11:25 -0400 Subject: [AI Seminar] Theory Seminar Course: Greatest Hits in Computer Science Theory 2016 Message-ID: <31e12b08-73f4-3dfc-eef8-8d4e31880490@cs.cmu.edu> Hi all, I am excited to teach the "Greatest Hits in Computer Science Theory 2016 " this semester. In this seminar PhD level course we will discuss the latest breakthroughs and greatest results in theoretical computer science published over the last two years in FOCS/STOC/SODA. The class is once a week Thursday from 12:00PM - 2:50PM. Have a look at the list of suggested papers . So many cool things to learn! If you are a PhD student interested in doing research in theory this might be your most efficient way to catch up on the the newest results and a great starting point for new research projects. I hope to see you in class. Cheers, Bernhard -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Thu Sep 1 08:18:39 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Thu, 1 Sep 2016 08:18:39 -0400 Subject: [AI Seminar] AI Lunch -- Bryan Hooi -- September 6th, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, September 6th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Bryan Hooi will give a talk titled "FRAUDAR: Bounding Graph Fraud in the Face of Camouflage." *Abstract:* Given a bipartite graph of users and the products that they review, or followers and followees, how can we detect fake reviews or follows? Existing fraud detection methods (spectral, etc.) try to identify dense subgraphs of nodes that are sparsely connected to the remaining graph. Fraudsters can evade these methods using camouflage, by adding reviews or follows with honest targets so that they look "normal". Even worse, some fraudsters use hijacked accounts from honest users, and then the camouflage is indeed organic. Our focus is to spot fraudsters in the presence of camouflage or hijacked accounts. We propose FRAUDAR, an algorithm that (a) is camouflage-resistant, (b) provides upper bounds on the effectiveness of fraudsters, and (c) is effective in real-world data. Experimental results under various attacks show that FRAUDAR outperforms the top competitor in accuracy of detecting both camouflaged and non-camouflaged fraud. Additionally, in real-world experiments with a Twitter follower-followee graph of 1.47 billion edges, FRAUDAR successfully detected a subgraph of more than 4000 detected accounts, of which a majority had tweets showing that they used follower-buying services. This is joint work with Hyun Ah Song, Alex Beutel, Neil Shah, Kijung Shin, and Christos Faloutsos and won the "best research paper" award in KDD 2016. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Sep 5 20:44:34 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 5 Sep 2016 20:44:34 -0400 Subject: [AI Seminar] Fwd: AI Lunch -- Bryan Hooi -- September 6th, 2016 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, September 6th. ---------- Forwarded message ---------- From: Ellen Vitercik Date: Thu, Sep 1, 2016 at 8:18 AM Subject: AI Lunch -- Bryan Hooi -- September 6th, 2016 To: bhooi at andrew.cmu.edu, ai-seminar-announce at cs.cmu.edu Dear faculty and students, We look forward to seeing you this Tuesday, September 6th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Bryan Hooi will give a talk titled "FRAUDAR: Bounding Graph Fraud in the Face of Camouflage." *Abstract:* Given a bipartite graph of users and the products that they review, or followers and followees, how can we detect fake reviews or follows? Existing fraud detection methods (spectral, etc.) try to identify dense subgraphs of nodes that are sparsely connected to the remaining graph. Fraudsters can evade these methods using camouflage, by adding reviews or follows with honest targets so that they look "normal". Even worse, some fraudsters use hijacked accounts from honest users, and then the camouflage is indeed organic. Our focus is to spot fraudsters in the presence of camouflage or hijacked accounts. We propose FRAUDAR, an algorithm that (a) is camouflage-resistant, (b) provides upper bounds on the effectiveness of fraudsters, and (c) is effective in real-world data. Experimental results under various attacks show that FRAUDAR outperforms the top competitor in accuracy of detecting both camouflaged and non-camouflaged fraud. Additionally, in real-world experiments with a Twitter follower-followee graph of 1.47 billion edges, FRAUDAR successfully detected a subgraph of more than 4000 detected accounts, of which a majority had tweets showing that they used follower-buying services. This is joint work with Hyun Ah Song, Alex Beutel, Neil Shah, Kijung Shin, and Christos Faloutsos and won the "best research paper" award in KDD 2016. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Sep 7 12:57:16 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 7 Sep 2016 12:57:16 -0400 Subject: [AI Seminar] AI Lunch -- Stephan Mandt -- September 13th, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, September 13th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . *We are looking for someone to give a talk on Tuesday, September 20th. *Please email me ( vitercik at cs.cmu.edu) and Ariel Procaccia (arielpro at cs.cmu.edu) if you would like to volunteer. On Tuesday, Stephan Mandt , a Research Scientist at Disney Research Pittsburgh, will give a talk titled "Variational Inference: From Artificial Temperatures to Stochastic Gradients." *Abstract:* Bayesian modeling is a popular approach to solving machine learning problems. In this talk, we will first review variational inference, where we map Bayesian inference to an optimization problem. This optimization problem is non-convex, meaning that there are many local optima that correspond to poor fits of the data. We first show that by introducing a ?local temperature? to every data point and applying the machinery of variational inference, we can avoid some of these poor optima, suppress the effects of outliers, and ultimately find more meaningful patterns. In the second part of the talk, we will then present a Bayesian view on Stochastic Gradient Descent (SGD). When operated with a constant, non-decreasing learning rates, SGD first marches towards the optimum of the objective and then samples from a stationary distribution that is centered around the optimum. As such, SGD resembles Markov Chain Monte Carlo (MCMC) algorithms which, after a burn-in period, draw samples from a Bayesian posterior. Drawing on the tools of variational inference, we investigate and formalize this connection. Our analysis reveals criteria that allow us to use SGD as an approximate scalable MCMC algorithm that can compete with more complicated state-of-the-art Bayesian approaches. *Speaker bio: *Stephan Mandt is a Research Scientist at Disney Research Pittsburgh where he leads the statistical machine learning group. Previously, he was a postdoctoral researcher with David Blei at Columbia University, where he worked on scalable approximate Bayesian inference algorithms. Trained as a statistical physicist, he held a previous postdoctoral fellowship at Princeton University and holds a Ph.D. from the University of Cologne as a fellow of the German National Merit Foundation. Personal website: www.stephanmandt.com Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Sep 12 08:59:08 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 12 Sep 2016 08:59:08 -0400 Subject: [AI Seminar] Fwd: AI Lunch -- Stephan Mandt -- September 13th, 2016 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, September 13th. ---------- Forwarded message ---------- From: Ellen Vitercik Date: Wed, Sep 7, 2016 at 12:57 PM Subject: AI Lunch -- Stephan Mandt -- September 13th, 2016 To: ai-seminar-announce at cs.cmu.edu, Stephan Mandt < stephan.mandt at disneyresearch.com> Dear faculty and students, We look forward to seeing you this Tuesday, September 13th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Stephan Mandt , a Research Scientist at Disney Research Pittsburgh, will give a talk titled "Variational Inference: From Artificial Temperatures to Stochastic Gradients." *Abstract:* Bayesian modeling is a popular approach to solving machine learning problems. In this talk, we will first review variational inference, where we map Bayesian inference to an optimization problem. This optimization problem is non-convex, meaning that there are many local optima that correspond to poor fits of the data. We first show that by introducing a ?local temperature? to every data point and applying the machinery of variational inference, we can avoid some of these poor optima, suppress the effects of outliers, and ultimately find more meaningful patterns. In the second part of the talk, we will then present a Bayesian view on Stochastic Gradient Descent (SGD). When operated with a constant, non-decreasing learning rates, SGD first marches towards the optimum of the objective and then samples from a stationary distribution that is centered around the optimum. As such, SGD resembles Markov Chain Monte Carlo (MCMC) algorithms which, after a burn-in period, draw samples from a Bayesian posterior. Drawing on the tools of variational inference, we investigate and formalize this connection. Our analysis reveals criteria that allow us to use SGD as an approximate scalable MCMC algorithm that can compete with more complicated state-of-the-art Bayesian approaches. *Speaker bio: *Stephan Mandt is a Research Scientist at Disney Research Pittsburgh where he leads the statistical machine learning group. Previously, he was a postdoctoral researcher with David Blei at Columbia University, where he worked on scalable approximate Bayesian inference algorithms. Trained as a statistical physicist, he held a previous postdoctoral fellowship at Princeton University and holds a Ph.D. from the University of Cologne as a fellow of the German National Merit Foundation. Personal website: www.stephanmandt.com Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Fri Sep 16 12:53:06 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Fri, 16 Sep 2016 12:53:06 -0400 Subject: [AI Seminar] AI Lunch -- Noam Brown -- September 20th, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, September 20th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Noam Brown will give a talk titled "Reduced Space and Faster Convergence in Imperfect-Information via Regret-Based Pruning." *Abstract*: Counterfactual Regret Minimization (CFR) is a leading algorithm for solving large zero-sum imperfect-information games. CFR is an iterative algorithm that repeatedly traverses the game tree, updating regrets at each information set. We introduce Regret-Based Pruning (RBP), an improvement to CFR that prunes any path of play in the tree, and its descendants, that has negative regret. It revisits that sequence at the earliest subsequent CFR iteration where the regret could have become positive, had that path been explored on every iteration. We prove that in zero-sum games it asymptotically prunes any action that is not part of a best response to some Nash equilibrium. This leads to provably faster convergence and lower space requirements. Experiments show that RBP can result in an order of magnitude reduction in both space and time needed to converge, and that the reduction factor increases with game size. In partial fulfillment of the Speaking Requirement. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Sep 19 05:10:25 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 19 Sep 2016 11:10:25 +0200 Subject: [AI Seminar] Fwd: AI Lunch -- Noam Brown -- September 20th, 2016 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, September 20th, at noon in NSH 3305. ---------- Forwarded message ---------- From: Ellen Vitercik Date: Fri, Sep 16, 2016 at 6:53 PM Subject: AI Lunch -- Noam Brown -- September 20th, 2016 To: ai-seminar-announce at cs.cmu.edu Dear faculty and students, We look forward to seeing you this Tuesday, September 20th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Noam Brown will give a talk titled "Reduced Space and Faster Convergence in Imperfect-Information Games via Regret-Based Pruning." *Abstract*: Counterfactual Regret Minimization (CFR) is a leading algorithm for solving large zero-sum imperfect-information games. CFR is an iterative algorithm that repeatedly traverses the game tree, updating regrets at each information set. We introduce Regret-Based Pruning (RBP), an improvement to CFR that prunes any path of play in the tree, and its descendants, that has negative regret. It revisits that sequence at the earliest subsequent CFR iteration where the regret could have become positive, had that path been explored on every iteration. We prove that in zero-sum games it asymptotically prunes any action that is not part of a best response to some Nash equilibrium. This leads to provably faster convergence and lower space requirements. Experiments show that RBP can result in an order of magnitude reduction in both space and time needed to converge, and that the reduction factor increases with game size. In partial fulfillment of the Speaking Requirement. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Fri Sep 23 17:11:34 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Fri, 23 Sep 2016 23:11:34 +0200 Subject: [AI Seminar] AI Lunch -- Ellen Vitercik -- September 27th, 2016 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, September 27th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, I (Ellen Vitercik ) will give a talk titled "Foundations of application-specific algorithm selection." *Abstract:* In many scientific domains, the underlying computational problems are frequently NP-hard, but a wealth of algorithms exist which efficiently find approximate solutions. These algorithms are often defined by a number of tunable parameters, which introduce an infinite spectrum of algorithm options. Which algorithm and choice of parameters will lead to the best performance for the specific application at hand? One could compare the worst-case performance of the algorithms in question and choose the algorithm with the best worst-case performance. However, worst-case instances may not occur in practice, which means that this technique will not offer guidance specific to the application at hand. This problem, often referred to as application-specific algorithm selection, has received significant attention from an empirical perspective over the past several decades, but it is not yet structured on a firm theoretical foundation. In this work, we develop learning algorithms for application-specific algorithm selection with robust guarantees. We focus on two large families of algorithms, the first of which is an infinite set of clustering algorithms based on hierarchical clustering followed by dynamic programming. The second family we analyze is a rich parameterized class of integer quadratic programming approximation algorithms based on SDP rounding. We work within a widely applicable framework wherein the learning algorithm is given samples from an application-specific distribution over problem instances and learns an algorithm that performs well over the distribution. We provide strong sample complexity guarantees and efficient algorithms which optimize the parameters to best suit typical inputs from a particular application. This is joint work with Nina Balcan, Vaishnavh Nagarajan, and Colin White. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Sep 26 10:09:45 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 26 Sep 2016 10:09:45 -0400 Subject: [AI Seminar] Fwd: AI Lunch -- Ellen Vitercik -- September 27th, 2016 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, September 27th, at noon in NSH 3305. ---------- Forwarded message ---------- From: Ellen Vitercik Date: Fri, Sep 23, 2016 at 5:11 PM Subject: AI Lunch -- Ellen Vitercik -- September 27th, 2016 To: ai-seminar-announce at cs.cmu.edu Dear faculty and students, We look forward to seeing you this Tuesday, September 27th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, I (Ellen Vitercik ) will give a talk titled "Foundations of application-specific algorithm selection." *Abstract:* In many scientific domains, the underlying computational problems are frequently NP-hard, but a wealth of algorithms exist which efficiently find approximate solutions. These algorithms are often defined by a number of tunable parameters, which introduce an infinite spectrum of algorithm options. Which algorithm and choice of parameters will lead to the best performance for the specific application at hand? One could compare the worst-case performance of the algorithms in question and choose the algorithm with the best worst-case performance. However, worst-case instances may not occur in practice, which means that this technique will not offer guidance specific to the application at hand. This problem, often referred to as application-specific algorithm selection, has received significant attention from an empirical perspective over the past several decades, but it is not yet structured on a firm theoretical foundation. In this work, we develop learning algorithms for application-specific algorithm selection with robust guarantees. We focus on two large families of algorithms, the first of which is an infinite set of clustering algorithms based on hierarchical clustering followed by dynamic programming. The second family we analyze is a rich parameterized class of integer quadratic programming approximation algorithms based on SDP rounding. We work within a widely applicable framework wherein the learning algorithm is given samples from an application-specific distribution over problem instances and learns an algorithm that performs well over the distribution. We provide strong sample complexity guarantees and efficient algorithms which optimize the parameters to best suit typical inputs from a particular application. This is joint work with Nina Balcan, Vaishnavh Nagarajan, and Colin White. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Thu Sep 29 10:13:12 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Thu, 29 Sep 2016 10:13:12 -0400 Subject: [AI Seminar] AI Lunch -- Room Change -- Jamie Morgenstern -- October 4 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, October 4th, at noon in *NSH 1507* for AI lunch. *Please note the room change*. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Jamie Morgenstern , a postdoc at UPenn and a CMU alum, will give a talk titled "Towards a Theory of Fairness in Machine Learning." *Abstract:* Algorithm design has moved from being a tool used exclusively for designing systems to one used to present people with personalized content, advertisements, and other economic opportunities. Massive amounts of information is recorded about people's online behavior including the websites they visit, the advertisements they click on, their search history, and their IP address. Algorithms then use this information for many purposes: to choose which prices to quote individuals for airline tickets, which advertisements to show them, and even which news stories to promote. These systems create new challenges for algorithm design. When a person's behavior influences the prices they may face in the future, they may have a strong incentive to modify their behavior to improve their long-term utility; therefore, these algorithms' performance should be resilient to strategic manipulation. Furthermore, when an algorithm makes choices that affect people's everyday lives, the effects of these choices raise ethical concerns such as whether the algorithm's behavior violates individuals' privacy or whether the algorithm treats people fairly. *Speaker bio:* Jamie Morgenstern is a Warren Center postdoctoral fellow in Computer Science and Economics at the University of Pennsylvania. She received her Ph.D. in Computer Science from Carnegie Mellon University in 2015, and her B.S. in Computer Science and B.A. in Mathematics from the University of Chicago in 2010. Her research focuses on machine learning for mechanism design, fairness in machine learning, and algorithmic game theory. She received a Microsoft Women's Research Scholarship, an NSF Graduate Research Fellowship, and a Simons Award for Graduate Students in Theoretical Computer Science. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Thu Sep 29 12:09:25 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Thu, 29 Sep 2016 12:09:25 -0400 Subject: [AI Seminar] AI Lunch -- Room Change -- Jamie Morgenstern -- October 4 In-Reply-To: References: Message-ID: I'm sorry, I accidentally left out part of the abstract. *Title*: Towards a Theory of Fairness in Machine Learning *Abstract: *Algorithm design has moved from being a tool used exclusively for designing systems to one used to present people with personalized content, advertisements, and other economic opportunities. Massive amounts of information is recorded about people's online behavior including the websites they visit, the advertisements they click on, their search history, and their IP address. Algorithms then use this information for many purposes: to choose which prices to quote individuals for airline tickets, which advertisements to show them, and even which news stories to promote. These systems create new challenges for algorithm design. When a person's behavior influences the prices they may face in thefuture, they may have a strong incentive to modify their behavior to improve their long-term utility; therefore, these algorithms' performance should be resilient to strategic manipulation. Furthermore, when an algorithm makes choices that affect people's everyday lives, the effects of these choices raise ethical concerns such as whether the algorithm's behavior violates individuals' privacy or whether the algorithm treats people fairly. Machine learning algorithms in particular have received much attention for exhibiting bias, or unfairness, in a large number of contexts. In this talk, I will describe my recent work on developing a definition of fairness for machine learning. One definition of fairness, encoding the notion of 'fair equality of opportunity', informally, states that if one person has higher expected quality than another person, the higher quality person should be given at least as much opportunity as the lower quality person. I will present a result characterizing the performance degradation of algorithms which satisfy this condition in the contextual bandits setting. To complement these theoretical results, I then present the results of several empirical evaluations of fair algorithms. I will also briefly describe my work on designing algorithms whose performance guarantees are resilient to strategic manipulation of their inputs, and machine learning for optimal auction design. *Speaker bio:* Jamie Morgenstern is a Warren Center postdoctoral fellow in Computer Science and Economics at the University of Pennsylvania. She received her Ph.D. in Computer Science from Carnegie Mellon University in 2015, and her B.S. in Computer Science and B.A. in Mathematics from the University of Chicago in 2010. Her research focuses on machine learning for mechanism design, fairness in machine learning, and algorithmic game theory. She received a Microsoft Women's Research Scholarship, an NSF Graduate Research Fellowship, and a Simons Award for Graduate Students in Theoretical Computer Science. On Thu, Sep 29, 2016 at 10:13 AM, Ellen Vitercik wrote: > Dear faculty and students, > > We look forward to seeing you this Tuesday, October 4th, at noon in *NSH > 1507* for AI lunch. *Please note the room change*. To learn more about > the seminar and lunch, please visit the AI Lunch webpage > . > > On Tuesday, Jamie Morgenstern , a > postdoc at UPenn and a CMU alum, will give a talk titled "Towards a > Theory of Fairness in Machine Learning." > > *Abstract:* Algorithm design has moved from being a tool used exclusively > for designing systems to one used to present people with personalized > content, advertisements, and other economic opportunities. Massive amounts > of information is recorded about people's online behavior including the > websites they visit, the advertisements they click on, their search > history, and their IP address. Algorithms then use this information for > many purposes: to choose which prices to quote individuals for airline > tickets, which advertisements to show them, and even which news stories to > promote. These systems create new challenges for algorithm design. When a > person's behavior influences the prices they may face in the future, they > may have a strong incentive to modify their behavior to improve their > long-term utility; therefore, these algorithms' performance should be > resilient to strategic manipulation. Furthermore, when an algorithm makes > choices that affect people's everyday lives, the effects of these choices > raise ethical concerns such as whether the algorithm's behavior violates > individuals' privacy or whether the algorithm treats people fairly. > > *Speaker bio:* Jamie Morgenstern is a Warren Center postdoctoral fellow > in Computer Science and Economics at the University of Pennsylvania. She > received her Ph.D. in Computer Science from Carnegie Mellon University in > 2015, and her B.S. in Computer Science and B.A. in Mathematics from the > University of Chicago in 2010. Her research focuses on machine learning for > mechanism design, fairness in machine learning, and algorithmic game > theory. She received a Microsoft Women's Research Scholarship, an NSF > Graduate Research Fellowship, and a Simons Award for Graduate Students in > Theoretical Computer Science. > > Best, > Ellen and Ariel > -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Oct 3 07:55:57 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 3 Oct 2016 07:55:57 -0400 Subject: [AI Seminar] AI Lunch -- Room Change -- Jamie Morgenstern -- October 4 Message-ID: Dear faculty and students, This is a reminder that AI lunch will be held tomorrow, Tuesday, October 4th, at noon in *NSH 1507. Please note the room change. *To learn more about the seminar and lunch, please visit the AI Lunch webpage . Jamie Morgenstern , a postdoc at UPenn and a CMU alum, will give a talk titled "Towards a Theory of Fairness in Machine Learning." *Abstract: *Algorithm design has moved from being a tool used exclusively for designing systems to one used to present people with personalized content, advertisements, and other economic opportunities. Massive amounts of information is recorded about people's online behavior including the websites they visit, the advertisements they click on, their search history, and their IP address. Algorithms then use this information for many purposes: to choose which prices to quote individuals for airline tickets, which advertisements to show them, and even which news stories to promote. These systems create new challenges for algorithm design. When a person's behavior influences the prices they may face in thefuture, they may have a strong incentive to modify their behavior to improve their long-term utility; therefore, these algorithms' performance should be resilient to strategic manipulation. Furthermore, when an algorithm makes choices that affect people's everyday lives, the effects of these choices raise ethical concerns such as whether the algorithm's behavior violates individuals' privacy or whether the algorithm treats people fairly. Machine learning algorithms in particular have received much attention for exhibiting bias, or unfairness, in a large number of contexts. In this talk, I will describe my recent work on developing a definition of fairness for machine learning. One definition of fairness, encoding the notion of 'fair equality of opportunity', informally, states that if one person has higher expected quality than another person, the higher quality person should be given at least as much opportunity as the lower quality person. I will present a result characterizing the performance degradation of algorithms which satisfy this condition in the contextual bandits setting. To complement these theoretical results, I then present the results of several empirical evaluations of fair algorithms. I will also briefly describe my work on designing algorithms whose performance guarantees are resilient to strategic manipulation of their inputs, and machine learning for optimal auction design. *Speaker bio:* Jamie Morgenstern is a Warren Center postdoctoral fellow in Computer Science and Economics at the University of Pennsylvania. She received her Ph.D. in Computer Science from Carnegie Mellon University in 2015, and her B.S. in Computer Science and B.A. in Mathematics from the University of Chicago in 2010. Her research focuses on machine learning for mechanism design, fairness in machine learning, and algorithmic game theory. She received a Microsoft Women's Research Scholarship, an NSF Graduate Research Fellowship, and a Simons Award for Graduate Students in Theoretical Computer Science. -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Oct 5 14:08:28 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 5 Oct 2016 14:08:28 -0400 Subject: [AI Seminar] AI Seminar -- Carla Gomes -- October 11 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, October 11th, at noon in NSH 3305 for AI Seminar. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . On Tuesday, Carla P. Gomes , Professor of Computer Science at Cornell University, will give a talk titled "Challenges for AI in Computational Sustainability." *Abstract: *Computational sustainability is a new interdisciplinary research field with the overarching goal of developing computational models, methods, and tools to help manage the balance between environmental, economic, and societal needs for a sustainable future. I will provide examples of computational sustainability problems, ranging from wildlife conservation and biodiversity, to poverty mitigation, to materials discovery for renewable energy materials. I will also highlight cross-cutting computational themes and challenges for AI at the intersection of constraint reasoning, optimization, machine learning, citizen science and crowd sourcing. *Bio:* Carla Gomes is a Professor of Computer Science at Cornell University, with joint appointments in the Dept. of Computer Science, Dept. of Information Science, and the Dyson School of Applied Economics and Management. Gomes obtained a Ph.D. in computer science in the area of artificial intelligence and operations research from the University of Edinburgh. Gomes?s central research themes are the integration of concepts from constraint and logical reasoning, mathematical programming, and machine learning, for large scale combinatorial problems; the study of the impact of structure on problem hardness; and the use of randomization techniques to improve the performance of search methods. More recently, Gomes has become deeply immersed in research in the new field of Computational Sustainability. From 2007-2013 Gomes led an NSF Expeditions-in-Computing in Computational Sustainability. Gomes and collaborators have successfully pioneered and nucleated the new field of Computational Sustainability. Gomes is currently the lead PI of a new NSF Expeditions-in-Computing that established CompSustNet, a large-scale national and international research network, to further expand the field and Computational Sustainability. Gomes is a Fellow of the Association for the Advancement of Artificial Intelligence (AAAI) and a Fellow of American Association for the Advancement of Science. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Oct 10 09:40:02 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 10 Oct 2016 09:40:02 -0400 Subject: [AI Seminar] Fwd: AI Seminar -- Carla Gomes -- October 11 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, October 11th, at noon in NSH 3305. ---------- Forwarded message ---------- From: Ellen Vitercik Date: Wed, Oct 5, 2016 at 2:08 PM Subject: AI Seminar -- Carla Gomes -- October 11 To: gomes at cs.cornell.edu, ai-seminar-announce at cs.cmu.edu Dear faculty and students, We look forward to seeing you this Tuesday, October 11th, at noon in NSH 3305 for AI Seminar. To learn more about the seminar and lunch, or to volunteer to give a talk, please visit the AI Lunch webpage . On Tuesday, Carla P. Gomes , Professor of Computer Science at Cornell University, will give a talk titled "Challenges for AI in Computational Sustainability." *Abstract: *Computational sustainability is a new interdisciplinary research field with the overarching goal of developing computational models, methods, and tools to help manage the balance between environmental, economic, and societal needs for a sustainable future. I will provide examples of computational sustainability problems, ranging from wildlife conservation and biodiversity, to poverty mitigation, to materials discovery for renewable energy materials. I will also highlight cross-cutting computational themes and challenges for AI at the intersection of constraint reasoning, optimization, machine learning, citizen science and crowd sourcing. *Bio:* Carla Gomes is a Professor of Computer Science at Cornell University, with joint appointments in the Dept. of Computer Science, Dept. of Information Science, and the Dyson School of Applied Economics and Management. Gomes obtained a Ph.D. in computer science in the area of artificial intelligence and operations research from the University of Edinburgh. Gomes?s central research themes are the integration of concepts from constraint and logical reasoning, mathematical programming, and machine learning, for large scale combinatorial problems; the study of the impact of structure on problem hardness; and the use of randomization techniques to improve the performance of search methods. More recently, Gomes has become deeply immersed in research in the new field of Computational Sustainability. From 2007-2013 Gomes led an NSF Expeditions-in-Computing in Computational Sustainability. Gomes and collaborators have successfully pioneered and nucleated the new field of Computational Sustainability. Gomes is currently the lead PI of a new NSF Expeditions-in-Computing that established CompSustNet, a large-scale national and international research network, to further expand the field and Computational Sustainability. Gomes is a Fellow of the Association for the Advancement of Artificial Intelligence (AAAI) and a Fellow of American Association for the Advancement of Science. Best, Ellen and Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From arielpro at cs.cmu.edu Tue Oct 11 21:58:44 2016 From: arielpro at cs.cmu.edu (Ariel Procaccia) Date: Tue, 11 Oct 2016 21:58:44 -0400 Subject: [AI Seminar] Can't (wait to) vote? Try RoboVote! Message-ID: Dear colleagues, For the past year and a half we've been working on RoboVote ( http://robovote.org/), a not-for-profit voting website that provides optimization-driven methods for making collective decisions. Check out RoboVote's about page for (somewhat AI-centric) information about the website, its algorithms, and the theory underlying them, as well as info about the amazing team. RoboVote will publicly launch in the week of Oct 24-28. In the meantime we're ironing out some last kinks, and *we ask for your help in testing the website*. Please try the demo or create an account and some actual polls. If you have any suggestions for improvement, encounter any bugs, or experience display issues (RoboVote is responsive and should look good on all devices), please submit your comments through RoboVote's feedback form. We would really appreciate it! Thanks, Ariel and Nisarg -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Oct 12 19:11:00 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 12 Oct 2016 19:11:00 -0400 Subject: [AI Seminar] AI Lunch -- Yisong Yue -- October 18 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, October 18th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Yisong Yue , a professor in the Computing and Mathematical Sciences department at the California Institute of Technology, will give a talk titled "The Dueling Bandits Problem." *Abstract: *In this talk, I will present the Dueling Bandits Problem, which is an online learning framework tailored towards real-time learning from subjective human feedback. In particular, the Dueling Bandits Problem only requires pairwise comparisons, which are shown to be reliably inferred in a variety of subjective feedback settings such as for information retrieval and recommender systems. I will provide an overview of the Dueling Bandits Problem with basic algorithmic results. I will then conclude by discussing some ongoing research directions with applications to personalized medicine. This is joint work with Josef Broder, Bobby Kleinberg, Thorsten Joachims, Yanan Sui, Vincent Zhuang, and Joel Burdick. -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Oct 17 09:04:54 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 17 Oct 2016 09:04:54 -0400 Subject: [AI Seminar] Fwd: AI Lunch -- Yisong Yue -- October 18 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, October 18th, at noon in NSH 3305. ---------- Forwarded message ---------- From: Ellen Vitercik Date: Wed, Oct 12, 2016 at 7:11 PM Subject: AI Lunch -- Yisong Yue -- October 18 To: ai-seminar-announce at cs.cmu.edu, Yisong Yue Dear faculty and students, We look forward to seeing you this Tuesday, October 18th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Yisong Yue , a professor in the Computing and Mathematical Sciences department at the California Institute of Technology, will give a talk titled "The Dueling Bandits Problem." *Abstract: *In this talk, I will present the Dueling Bandits Problem, which is an online learning framework tailored towards real-time learning from subjective human feedback. In particular, the Dueling Bandits Problem only requires pairwise comparisons, which are shown to be reliably inferred in a variety of subjective feedback settings such as for information retrieval and recommender systems. I will provide an overview of the Dueling Bandits Problem with basic algorithmic results. I will then conclude by discussing some ongoing research directions with applications to personalized medicine. This is joint work with Josef Broder, Bobby Kleinberg, Thorsten Joachims, Yanan Sui, Vincent Zhuang, and Joel Burdick. -------------- next part -------------- An HTML attachment was scrubbed... URL: From arielpro at cs.cmu.edu Tue Oct 18 17:10:32 2016 From: arielpro at cs.cmu.edu (Ariel Procaccia) Date: Tue, 18 Oct 2016 17:10:32 -0400 Subject: [AI Seminar] Talk by Ashish Goel -- October 25 (+ sign up for meetings!) Message-ID: Hi all, Ashish Goel (Stanford) will give an AI lunch talk on October 25 at noon in NSH 3305, entitled "Decision Making at Scale: Algorithms and Deployments". You can find the abstract here . Lunch from Sukhothai will be served! *Ashish will be available for meetings on October 25.* If you'd like to meet him, please send me your constraints by replying to this email. Best, Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From pdonti at cs.cmu.edu Fri Oct 21 16:40:47 2016 From: pdonti at cs.cmu.edu (Priya Donti) Date: Fri, 21 Oct 2016 16:40:47 -0400 Subject: [AI Seminar] Computational Sustainability Network Virtual Seminar (October 25, 4-5pm) Message-ID: Dear faculty and students, As Carla Gomes mentioned in her talk last week, she is one of the founders of CompSustNet , a network of researchers working at the intersection of computation and environmental sustainability. This network hosts a computational sustainability virtual seminar series every few weeks, the next installment of which is this upcoming Tuesday, October 25 (details below). We hope you'll join for all or part of this special seminar series! We're additionally launching a new mailing list for computational sustainability at CMU (https://mailman.srv.cs.cmu.edu/mailman/listinfo/comp- sustainability). Please consider joining if you would like to keep updated on research and happenings at the intersection of computation and environmental sustainability/related societal applications. Seminar details: *CompSustNet Seminar Tuesday, October 25, 4-5pm EDT *(register here ): *Information below from the CompSustNet website.* *Speakers:* Milind Tambe and Eric Rice, University of Southern California *Title:* How Can AI be Used for Social Good? Key Techniques, Applications, and Results *Abstract:* Discussions about the future negative consequences of AI sometimes drown out discussions of the current accomplishments and future potential of AI in helping us solve complex societal problems. At the USC Center for AI in Society , CAIS, our focus is on exploring AI research in tackling wicked problems in society. This talk will highlight the goals of CAIS and three areas of ongoing work. First, we will focus on the use of AI for assisting low-resource sections of society, such as homeless youth. Harnessing the social networks of such youth, we will illustrate the use of AI algorithms to help more effectively spread health information, such as for reducing risk of HIV infections. These algorithms have been piloted in homeless shelters in Los Angeles, and have shown significant improvements over traditional methods. Second, we will outline the use of AI for protection of forests, fish, and wildlife; learning models of adversary behavior allows us to predict poaching activities and plan effective patrols to deter them. These algorithms are in use in multiple countries, and we discuss concrete results we have obtained in a national park in Uganda. Finally, we will focus on the challenge of AI for public safety and security, discussing game theoretic algorithms for effective security resource allocation that are in actual daily use by agencies such as the US Coast Guard and the Federal Air Marshals Service to assist the protection of ports, airports, flights, and other critical infrastructure. These are just a few of the projects at CAIS, and we expect these and future projects at CAIS to continue to illustrate the significant potential that AI has for social good. *Bio:* Milind Tambe is Founding Co-Director of CAIS, the USC Center for AI for Society, and Helen N. and Emmett H. Jones Professor in Engineering at the University of Southern California(USC). He is a fellow of AAAI and ACM, as well as recipient of the ACM/SIGART Autonomous Agents Research Award, Christopher Columbus Fellowship Foundation Homeland security award, INFORMS Wagner prize for excellence in Operations Research practice, Rist Prize of the Military Operations Research Society, IBM Faculty Award, Okawa foundation faculty research award, RoboCup scientific challenge award, and other local awards such as the Orange County Engineering Council Outstanding Project Achievement Award, USC Associates award for creativity in research and USC Viterbi use-inspired research award. Prof. Tambe has contributed several foundational papers in AI in areas such as multiagent teamwork, distributed constraint optimization (DCOP) and security games. For this research, he has received the "influential paper award" and a number of best paper awards at conferences such as AAMAS, IJCAI, IAAI and IVA. In addition, Prof. Tambe pioneering real-world deployments of "security games" has led him and his team to receive the US Coast Guard Meritorious Team Commendation from the Commandant, US Coast Guard First District's Operational Excellence Award, Certificate of Appreciation from the US Federal Air Marshals Service and special commendation given by LA Airport police from the city of Los Angeles. For his teaching and service, Prof. Tambe has received the USC Steven B. Sample Teaching and Mentoring award and the ACM recognition of service award. He has also co-founded a company based on his research, Avata Intelligence, where he serves as the director of research. Prof. Tambe received his Ph.D. from the School of Computer Science at Carnegie Mellon University. Thanks and best regards, Priya ------------------------------------------------------------------ *Priya L. Donti* *Ph.D. Student* Department of Computer Science Department of Engineering and Public Policy Carnegie Mellon University ------------------------------------------------------------------ On Wed, Oct 5, 2016 at 2:08 PM, Ellen Vitercik wrote: > Dear faculty and students, > > We look forward to seeing you this Tuesday, October 11th, at noon in NSH > 3305 for AI Seminar. To learn more about the seminar and lunch, or to > volunteer to give a talk, please visit the AI Lunch webpage > . > > On Tuesday, Carla P. Gomes , Professor > of Computer Science at Cornell University, will give a talk titled > "Challenges for AI in Computational Sustainability." > > *Abstract: *Computational sustainability is a new interdisciplinary > research field with the overarching goal of developing computational > models, methods, and tools to help manage the balance between > environmental, economic, and societal needs for a sustainable future. I > will provide examples of computational sustainability problems, ranging > from wildlife conservation and biodiversity, to poverty mitigation, to > materials discovery for renewable energy materials. I will also highlight > cross-cutting computational themes and challenges for AI at the > intersection of constraint reasoning, optimization, machine learning, > citizen science and crowd sourcing. > > *Bio:* Carla Gomes is a Professor of Computer Science at Cornell > University, with joint appointments in the Dept. of Computer Science, Dept. > of Information Science, and the Dyson School of Applied Economics and > Management. Gomes obtained a Ph.D. in computer science in the area of > artificial intelligence and operations research from the University of > Edinburgh. Gomes?s central research themes are the integration of concepts > from constraint and logical reasoning, mathematical programming, and > machine learning, for large scale combinatorial problems; the study of the > impact of structure on problem hardness; and the use of randomization > techniques to improve the performance of search methods. More recently, > Gomes has become deeply immersed in research in the new field of > Computational Sustainability. From 2007-2013 Gomes led an NSF > Expeditions-in-Computing in Computational Sustainability. Gomes and > collaborators have successfully pioneered and nucleated the new field of > Computational Sustainability. Gomes is currently the lead PI of a new NSF > Expeditions-in-Computing that established CompSustNet, a large-scale > national and international research network, to further expand the field > and Computational Sustainability. Gomes is a Fellow of the Association for > the Advancement of Artificial Intelligence (AAAI) and a Fellow of American > Association for the Advancement of Science. > > Best, > > Ellen and Ariel > -------------- next part -------------- An HTML attachment was scrubbed... URL: From arielpro at cs.cmu.edu Tue Oct 25 08:39:06 2016 From: arielpro at cs.cmu.edu (Ariel Procaccia) Date: Tue, 25 Oct 2016 08:39:06 -0400 Subject: [AI Seminar] Fwd: Talk by Ashish Goel -- October 25 (+ sign up for meetings!) In-Reply-To: References: Message-ID: Reminder: this talk is today at noon. Ariel ---------- Forwarded message ---------- From: Ariel Procaccia Date: Tue, Oct 18, 2016 at 5:10 PM Subject: Talk by Ashish Goel -- October 25 (+ sign up for meetings!) To: ai-seminar-announce at cs.cmu.edu Hi all, Ashish Goel (Stanford) will give an AI lunch talk on October 25 at noon in NSH 3305, entitled "Decision Making at Scale: Algorithms and Deployments". You can find the abstract here . Lunch from Sukhothai will be served! *Ashish will be available for meetings on October 25.* If you'd like to meet him, please send me your constraints by replying to this email. Best, Ariel -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Oct 26 17:26:25 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 26 Oct 2016 17:26:25 -0400 Subject: [AI Seminar] AI Lunch -- Hadi Hosseini -- November 1 Message-ID: Hi all, We look forward to seeing you this Tuesday, November 1st, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Hadi Hosseini will give a talk titled "Analyzing and Designing Truthful Matching Mechanisms." *Abstract:* The problem of allocating indivisible goods to a set of self-interested agents in the absence of transferable utilities (such as money) is omnipresent in various resource allocation settings such as assigning shifts to nurses, dormitory rooms to students, and members to subcommittees. These settings leverage techniques from computer science and economics to ensure fairness and efficiency while preventing agents from manipulating the outcomes. In the first part of this talk, I will focus on two widely-studied randomized matching mechanisms for fair allocation of indivisible goods under ordinal preferences, namely Random Serial Dictatorship and Probabilistic Serial rule. I will give an overview of their properties and discuss how empirical results can provide deeper insights into theoretical guarantees, addressing the question of which mechanism to adopt in practice. In the second part, I will focus on sequential matching with dynamic ordinal preferences. I will briefly describe a novel model based on a generic stochastic decision process and show that, in contrast to static settings, traditional approaches are highly susceptible to manipulation in dynamic settings. I will describe how we can restore some of the desired properties by careful consideration of the history of outcomes and how this history-dependent approach impacts efficiency and fairness. This talk is based on joint work with Kate Larson and Robin Cohen. -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Oct 31 17:45:13 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 31 Oct 2016 17:45:13 -0400 Subject: [AI Seminar] Fwd: AI Lunch -- Hadi Hosseini -- November 1 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, November 1st, at noon in NSH 3305. ---------- Forwarded message ---------- From: Ellen Vitercik Date: Wed, Oct 26, 2016 at 5:26 PM Subject: AI Lunch -- Hadi Hosseini -- November 1 To: ai-seminar-announce at cs.cmu.edu, Hadi Hosseini < hadi.hosseini at uwaterloo.ca> Hi all, We look forward to seeing you this Tuesday, November 1st, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Hadi Hosseini will give a talk titled "Analyzing and Designing Truthful Matching Mechanisms." *Abstract:* The problem of allocating indivisible goods to a set of self-interested agents in the absence of transferable utilities (such as money) is omnipresent in various resource allocation settings such as assigning shifts to nurses, dormitory rooms to students, and members to subcommittees. These settings leverage techniques from computer science and economics to ensure fairness and efficiency while preventing agents from manipulating the outcomes. In the first part of this talk, I will focus on two widely-studied randomized matching mechanisms for fair allocation of indivisible goods under ordinal preferences, namely Random Serial Dictatorship and Probabilistic Serial rule. I will give an overview of their properties and discuss how empirical results can provide deeper insights into theoretical guarantees, addressing the question of which mechanism to adopt in practice. In the second part, I will focus on sequential matching with dynamic ordinal preferences. I will briefly describe a novel model based on a generic stochastic decision process and show that, in contrast to static settings, traditional approaches are highly susceptible to manipulation in dynamic settings. I will describe how we can restore some of the desired properties by careful consideration of the history of outcomes and how this history-dependent approach impacts efficiency and fairness. This talk is based on joint work with Kate Larson and Robin Cohen. -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Fri Nov 4 09:55:18 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Fri, 4 Nov 2016 09:55:18 -0400 Subject: [AI Seminar] AI Lunch -- Zhaohan (Daniel) Guo -- November 8 Message-ID: Hi all, We look forward to seeing you this Tuesday, November 8th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Zhaohan (Daniel) Guo will give a talk titled "A PAC RL Algorithm for Episodic POMDPs." *Abstract:* Many interesting real world domains involve reinforcement learning (RL) in partially observable environments. Efficient learning in such domains is important, but existing sample complexity bounds for partially observable RL are at least exponential in the episode length. Polynomial sample complexity bounds are prevalent for fully observable environments, so we looked for a way to do the same for POMDPs. Generally, polynomial sample complexity bounds for POMDPs are impossible, since observations may give no information about the underlying dynamics. However we can build on recent advances in estimating latent variable models using method of moments, specifically confidence bounds on method of moment estimators for HMMs. These methods quantify problem specific properties, allowing us to give, to our knowledge, the first partially observable RL algorithm with a problem specific polynomial bound on the sample complexity. -------------- next part -------------- An HTML attachment was scrubbed... URL: From arielpro at cs.cmu.edu Fri Nov 4 13:04:27 2016 From: arielpro at cs.cmu.edu (Ariel Procaccia) Date: Fri, 4 Nov 2016 13:04:27 -0400 Subject: [AI Seminar] Fwd: OR Seminar on Friday, November 11, 2016 In-Reply-To: References: <75ADA34501D6499DACE04A9AB3A357BB@tepper.cmu.edu> Message-ID: Talk of possible interest. Ariel ---------- Forwarded message ---------- From: Willem-Jan van Hoeve Date: Fri, Nov 4, 2016 at 12:38 PM Subject: Fwd: OR Seminar on Friday, November 11, 2016 To: arielpro at cs.cmu.edu Hi Ariel, Next week's OR seminar may be of interest to attendants of the AI seminar as well. Could you please forward the announcement via the AI Lunch Seminar mailinglist? Thanks, Willem -------- Forwarded Message -------- Subject: OR Seminar on Friday, November 11, 2016 Date: Fri, 4 Nov 2016 08:21:23 -0400 From: maobrien at andrew.cmu.edu To: vanhoeve at andrew.cmu.edu **Distributed via the Faculty Services mail distribution system.** The below OR seminar is posted at: https://econ.tepper.cmu.edu/Se minars/seminar.asp? dbaction=y&sort=1&short=Y&Restrict=All&Seminar+Area=Organization+Behavi or&keyword= Please take the time to schedule a meeting with the speaker on an individual basis When you are at the seminar site, proceed to this seminar. To add yourself for a meeting, click on View/Edit Schedule link and then click on the Edit Schedule link. Enter the name portion of your e-mail address (the @andrew.cmu.edu part is not needed) and click Update at bottom of page. Name: Dimitris Bertsimas, MIT Date: Friday, November 11, 2016 Time: 1:30 to 3:00 pm Location: Faculty Conference Room 322 Title: Machine learning and statistics via a modern optimization lens Abstract: The field of Statistics has historically been linked with Probability Theory. However, some of the central problems of classification, regression and estimation can naturally be written as optimization problems. While continuous optimization approaches has had a significant impact in Statistics, mixed integer optimization (MIO) has played a very limited role, primarily based on the belief that MIO models are computationally intractable. The period 1991-2015 has witnessed a) algorithmic advances in mixed integer optimization (MIO), which coupled with hardware improvements have resulted in an astonishing 450 billion factor speedup in solving MIO problems, b) significant advances in our ability to model and solve very high dimensional robust and convex optimization models. In this talk, we demonstrate that modern convex, robust and especially mixed integer optimization methods, when applied to a variety of classical Machine Learning (ML) /Statistics (S) problems can lead to certifiable optimal solutions for large scale instances that have often significantly improved out of sample accuracy compared to heuristic methods used in ML/S. Specifically, we report results on 1) The classical variable selection problem in regression currently solved by Lasso heuristically. 2) We show that robustness and not sparsity is the major reason of the success of Lasso in contrast to widely held beliefs in ML/S. 3) A systematic approach to design linear and logistic regression models based on MIO. 4) Optimal trees for classification solved by CART heuristically. 5) Robust classification including robust Logistic regression, robust optimal trees and robust support vector machines. 6) Sparse matrix estimation problems: Principal Component Analysis, Factor Analysis and Covariance matrix estimation. In all cases we demonstrate that optimal solutions to large scale instances (a) can be found in seconds, (b) can be certified to be optimal in minutes and (c) outperform classical approaches. Most importantly, this body of work suggests that linking ML/S to modern optimization leads to significant advances. -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Nov 7 08:16:31 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 7 Nov 2016 08:16:31 -0500 Subject: [AI Seminar] Fwd: AI Lunch -- Zhaohan (Daniel) Guo -- November 8 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, November 8th, at noon in NSH 3305. ---------- Forwarded message ---------- From: Ellen Vitercik Date: Fri, Nov 4, 2016 at 9:55 AM Subject: AI Lunch -- Zhaohan (Daniel) Guo -- November 8 To: ai-seminar-announce at cs.cmu.edu, "Zhaohan (Daniel) Guo" Hi all, We look forward to seeing you this Tuesday, November 8th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Zhaohan (Daniel) Guo will give a talk titled "A PAC RL Algorithm for Episodic POMDPs." *Abstract:* Many interesting real world domains involve reinforcement learning (RL) in partially observable environments. Efficient learning in such domains is important, but existing sample complexity bounds for partially observable RL are at least exponential in the episode length. Polynomial sample complexity bounds are prevalent for fully observable environments, so we looked for a way to do the same for POMDPs. Generally, polynomial sample complexity bounds for POMDPs are impossible, since observations may give no information about the underlying dynamics. However we can build on recent advances in estimating latent variable models using method of moments, specifically confidence bounds on method of moment estimators for HMMs. These methods quantify problem specific properties, allowing us to give, to our knowledge, the first partially observable RL algorithm with a problem specific polynomial bound on the sample complexity. -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Nov 9 14:11:26 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 9 Nov 2016 14:11:26 -0500 Subject: [AI Seminar] AI Lunch -- Andrew Moore -- November 15 Message-ID: Hi all, We look forward to seeing you this Tuesday, November 15th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Andrew Moore , Dean of the SCS at CMU, will give a talk. The talk description is below. TOKeN: The Open Knowledge Network: Creating the Semantic Information Infrastructure for the Future RV Guha, Schema.org; Andrew Moore (Presenting), Carnegie Mellon University Motivation Natural interfaces to large knowledge structures have the potential to impact science, education and business to an extent comparable to the WWW. We are already seeing the first wave of this in consumer services such as Siri, Cortana and Alexa. But these services are limited in their scope of knowledge, not open to direct access or contributors beyond their corporate firewalls, and can only answer relatively limited questions in their business areas. We now have the technology and know how to expand to thousands of new topic areas and many more useful classes of questions, if we mount an open effort to build a national or international knowledge graph. The architecture should allow people to encode knowledge for their topics of interest and be able to hook them into the larger network, without having to go through gatekeepers (such as Google or Apple). Once this knowledge is encoded, access to this should not be restricted to a small priesthood of SQL or other programmatic interface users. There will be a wide range of interfaces, including natural language interfaces, graphical interfaces and visualizations which no one has even invented yet. Developers will be able to independently create more sophisticated programs for answering queries, providing summaries that help regular people make decisions in their lives. This talk will summarize a discussion between a set of academics, internet companies and government agencies and go through the questions of "why now", "haven't we all tried this before", "what are the first steps the nation could take here", and "what exactly is it that we're proposing here?" This is not fully baked and so I will leave plenty of time for feedback and discussion. -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Fri Nov 11 17:32:20 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Fri, 11 Nov 2016 17:32:20 -0500 Subject: [AI Seminar] AI Lunch -- Call for student coordinator Message-ID: Hi everyone, I have now been the AI Lunch and Seminar student coordinator for almost a year, so I am reaching out to see if any current students are interested in taking over this January as student coordinator for the second year of AI lunch. I highly recommend it! It's a great and easy way to be involved with the AI group at CMU as a whole. Please contact me if you are interested or have any questions. Best, Ellen Vitercik -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Nov 14 11:18:32 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 14 Nov 2016 11:18:32 -0500 Subject: [AI Seminar] AI Lunch -- Andrew Moore -- November 15 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, November 15th, at noon in NSH 3305. On Wed, Nov 9, 2016 at 2:11 PM, Ellen Vitercik wrote: > Hi all, > > We look forward to seeing you this Tuesday, November 15th, at noon in NSH > 3305 for AI lunch. To learn more about the seminar and lunch, please visit > the AI Lunch webpage . > > On Tuesday, Andrew Moore , Dean of the SCS > at CMU, will give a talk. The talk description is below. > > TOKeN: The Open Knowledge Network: > > Creating the Semantic Information Infrastructure for the Future > > RV Guha, Schema.org; Andrew Moore (Presenting), Carnegie Mellon University > > Motivation > > Natural interfaces to large knowledge structures have the potential to > impact science, education and business to an extent comparable to the WWW. > We are already seeing the first wave of this in consumer services such as > Siri, Cortana and Alexa. But these services are limited in their scope of > knowledge, not open to direct access or contributors beyond their corporate > firewalls, and can only answer relatively limited questions in their > business areas. We now have the technology and know how to expand to > thousands of new topic areas and many more useful classes of questions, if > we mount an open effort to build a national or international knowledge > graph. > > The architecture should allow people to encode knowledge for their topics > of interest and be able to hook them into the larger network, without > having to go through gatekeepers (such as Google or Apple). > > Once this knowledge is encoded, access to this should not be restricted to > a small priesthood of SQL or other programmatic interface users. There will > be a wide range of interfaces, including natural language interfaces, > graphical interfaces and visualizations which no one has even invented > yet. Developers will be able to independently create more sophisticated > programs for answering queries, providing summaries that help regular > people make decisions in their lives. > > This talk will summarize a discussion between a set of academics, internet > companies and government agencies and go through the questions of "why > now", "haven't we all tried this before", "what are the first steps the > nation could take here", and "what exactly is it that we're proposing here?" > > > This is not fully baked and so I will leave plenty of time for feedback > and discussion. > -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Wed Nov 16 20:27:49 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Wed, 16 Nov 2016 20:27:49 -0500 Subject: [AI Seminar] AI Lunch -- Ariel Procaccia -- November 22 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, November 22nd, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Ariel Procaccia will give a talk titled "Computational Social Choice: For the People." *Abstract:* Computational social choice deals with algorithms for aggregating individual preferences or opinions towards collective decisions. AI researchers (including myself) have long argued that such algorithms could play a crucial role in the design and implementation of multiagent systems. However, in the last few years I have come to realize that the "killer app" of computational social choice is helping people -- not software agents -- make joint decisions. I will illustrate this theme through two recent endeavors: Spliddit.org, a website that offers provably fair solutions to everyday problems; and Robovote.org, which provides optimization-driven voting methods. Throughout the talk, I will devote special attention to the theoretical foundations and results that make these services possible. -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Sat Nov 19 16:10:17 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Sat, 19 Nov 2016 16:10:17 -0500 Subject: [AI Seminar] AI Lunch -- Call for Talks Spring 2017 Message-ID: Dear faculty and students, As this semester wraps us, we are beginning to look ahead to next semester's AI lunch schedule. If you are interested in giving a talk, I encourage you to contact me and Ariel (vitercik at cs.cmu.edu, arielpro at cs.cmu.edu) soon before the slots fill up. We already have a few talks scheduled (see the schedule here ), and the first available date is January 31st. As a reminder, the talks are every Tuesday at noon. It's been great to see what a success the first year of AI lunch proved to be, and we trust that you will help make this a lasting tradition at CMU! Best, Ellen -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Nov 21 12:11:52 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 21 Nov 2016 12:11:52 -0500 Subject: [AI Seminar] Fwd: AI Lunch -- Ariel Procaccia -- November 22 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, November 22nd, at noon in NSH 3305. ---------- Forwarded message ---------- From: Ellen Vitercik Date: Wed, Nov 16, 2016 at 8:27 PM Subject: AI Lunch -- Ariel Procaccia -- November 22 To: ai-seminar-announce at cs.cmu.edu, Ariel Procaccia Dear faculty and students, We look forward to seeing you this Tuesday, November 22nd, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Ariel Procaccia will give a talk titled "Computational Social Choice: For the People." *Abstract:* Computational social choice deals with algorithms for aggregating individual preferences or opinions towards collective decisions. AI researchers (including myself) have long argued that such algorithms could play a crucial role in the design and implementation of multiagent systems. However, in the last few years I have come to realize that the "killer app" of computational social choice is helping people -- not software agents -- make joint decisions. I will illustrate this theme through two recent endeavors: Spliddit.org, a website that offers provably fair solutions to everyday problems; and Robovote.org, which provides optimization-driven voting methods. Throughout the talk, I will devote special attention to the theoretical foundations and results that make these services possible. -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Sun Nov 27 09:20:38 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Sun, 27 Nov 2016 09:20:38 -0500 Subject: [AI Seminar] AI Lunch -- Juan Pablo Mendoza -- November 29 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, November 29th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Juan Pablo Mendoza will give a talk titled "Detection of Subtle and Context-Dependent Robot Model Inaccuracies." *Abstract:* Autonomous robots frequently rely on models of their sensing and actions for intelligent decision-making. Unfortunately, in complex environments, robots are bound to encounter situations in which their models do not accurately represent the world. Furthermore, these context-dependent model inaccuracies may be subtle, such that multiple observations may be necessary to distinguish them from process noise. We explore the problem of detection and correction of such subtle contextual model inaccuracies in high-dimensional autonomous robot domains. Our solution relies on reasoning about these contextual inaccuracies as parametric Regions of Inaccurate Modeling (RIMs) in the robot?s context space, and developing optimization and search-based algorithms for finding these RIMs. We describe the solution in detail, and explore its application to model inaccuracy detection in the CoBot mobile service robots and the CMDragons autonomous soccer robot team. -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Thu Dec 1 08:58:20 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Thu, 1 Dec 2016 08:58:20 -0500 Subject: [AI Seminar] AI Lunch -- Christian Kroer -- December 6 Message-ID: Dear faculty and students, We look forward to seeing you this Tuesday, December 6th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Christian Kroer will give a talk titled "First-Order Methods for Extensive-Form Game Solving." *Abstract:* We study the problem of computing a Nash equilibrium in large-scale two-player zero-sum extensive-form games. While this problem can be solved in polynomial time, sparse iterative methods, in particular regret- based methods and first-order methods (FOMs), are much cheaper and effective and hence usually preferred for large games. Thus far, regret-based methods have largely been favored in practice in spite of their theoretically inferior convergence rates. In this talk we investigate the acceleration of FOMs both theoretically and numerically. The convergence rates of FOMs depend heavily on the properties of the distance-generating function (DGF) that they are based on. We investigate the acceleration of FOMs for solving extensive-form games through a better design of the dilated entropy function--a class of DGFs related to the domains associated with the extensive-form games--and suggesting new sampling schemes for stochastic FOMs. Our new weighting scheme for the dilated entropy function establishes the first bound on the strong-convexity parameter for DGFs over general treeplexes, i.e., the strategy spaces of general sequential games, and it has no dependence on the branching factor of the player. Thus, we establish first explicit FOM convergence rates for general treeplexes, and significantly improve upon previous results given only for some special cases. Building on this result, we also introduce a class of gradient estimators which, along with our DGF, leads to the first stochastic FOM for extensive-form games. Experimentally, we investigate the performance of three FOMs (the excessive gap technique, mirror prox, and stochastic mirror prox) and compare their performance to the regret-based algorithms. Equipped with our distance-generating function, we find that mirror prox and the excessive gap technique outperform the prior regret-based methods for finding medium accuracy solutions. Joint work with Kevin Waugh, Fatma Kilinc-Karzan, and Tuomas Sandholm. -------------- next part -------------- An HTML attachment was scrubbed... URL: From vitercik at cs.cmu.edu Mon Dec 5 08:59:22 2016 From: vitercik at cs.cmu.edu (Ellen Vitercik) Date: Mon, 5 Dec 2016 08:59:22 -0500 Subject: [AI Seminar] Fwd: AI Lunch -- Christian Kroer -- December 6 In-Reply-To: References: Message-ID: This is a reminder that this talk is tomorrow, Tuesday, December 6th, at noon in NSH 3305. Also, we have recently uploaded Andrew Moore's great talk from November 15th to the new AI lunch Youtube channel. We intend to update the channel regularly from here on out. See the talk here . ---------- Forwarded message ---------- From: Ellen Vitercik Date: Thu, Dec 1, 2016 at 8:58 AM Subject: AI Lunch -- Christian Kroer -- December 6 To: ai-seminar-announce at cs.cmu.edu, Christian Kroer Dear faculty and students, We look forward to seeing you this Tuesday, December 6th, at noon in NSH 3305 for AI lunch. To learn more about the seminar and lunch, please visit the AI Lunch webpage . On Tuesday, Christian Kroer will give a talk titled "First-Order Methods for Extensive-Form Game Solving." *Abstract:* We study the problem of computing a Nash equilibrium in large-scale two-player zero-sum extensive-form games. While this problem can be solved in polynomial time, sparse iterative methods, in particular regret- based methods and first-order methods (FOMs), are much cheaper and effective and hence usually preferred for large games. Thus far, regret-based methods have largely been favored in practice in spite of their theoretically inferior convergence rates. In this talk we investigate the acceleration of FOMs both theoretically and numerically. The convergence rates of FOMs depend heavily on the properties of the distance-generating function (DGF) that they are based on. We investigate the acceleration of FOMs for solving extensive-form games through a better design of the dilated entropy function--a class of DGFs related to the domains associated with the extensive-form games--and suggesting new sampling schemes for stochastic FOMs. Our new weighting scheme for the dilated entropy function establishes the first bound on the strong-convexity parameter for DGFs over general treeplexes, i.e., the strategy spaces of general sequential games, and it has no dependence on the branching factor of the player. Thus, we establish first explicit FOM convergence rates for general treeplexes, and significantly improve upon previous results given only for some special cases. Building on this result, we also introduce a class of gradient estimators which, along with our DGF, leads to the first stochastic FOM for extensive-form games. Experimentally, we investigate the performance of three FOMs (the excessive gap technique, mirror prox, and stochastic mirror prox) and compare their performance to the regret-based algorithms. Equipped with our distance-generating function, we find that mirror prox and the excessive gap technique outperform the prior regret-based methods for finding medium accuracy solutions. Joint work with Kevin Waugh, Fatma Kilinc-Karzan, and Tuomas Sandholm. -------------- next part -------------- An HTML attachment was scrubbed... URL: From sandholm at cs.cmu.edu Fri Dec 9 20:00:23 2016 From: sandholm at cs.cmu.edu (Tuomas Sandholm) Date: Fri, 9 Dec 2016 20:00:23 -0500 Subject: [AI Seminar] FW: Paul Milgrom In-Reply-To: References: Message-ID: Talk of interest to some of you.. Best, Tuomas From: Alexey Kushnir [mailto:alexey.kushnir at gmail.com] Sent: Friday, December 09, 2016 2:08 PM To: Tuomas Sandholm Subject: Paul Milgrom Dear Tuomas, I want to attract your attention to the next Tepper microeconomics seminar speaker, who might be of your interest. Paul Milgrom will be presenting next Thursday December 15th at 4:00-5:30 in Tepper/GSIA Faculty Conference Room 322. This is actually his first time visiting CMU! Feel free to pass this information to anybody who might be interested. Let me also know if you want to join lunch or dinner with him. Best, Alex -- Alexey Kushnir, Assistant Professor of Economics, Tepper School of Business, Carnegie Mellon University, http://www.andrew.cmu.edu/user/akushnir/ -------------- next part -------------- An HTML attachment was scrubbed... URL: