- talks - lists.cs.princeton.edu

Kaifeng Lyu will present his General Exam "Theoretical Analysis of Modern Neural Networks with Normalization Layers: Auto Rate-Tuning and Sharpness Reduction" on Tuesday, January 17, 2023 at 1:00 PM over Zoom.
by CS Grad Department 13 Jan '23

13 Jan '23

Kaifeng Lyu will present his General Exam "Theoretical Analysis of Modern Neural Networks with Normalization Layers: Auto Rate-Tuning and Sharpness Reduction" on Tuesday, January 17, 2023 at 1:00 PM over Zoom. Zoom link: [ https://princeton.zoom.us/j/97584410651 | https://princeton.zoom.us/j/97584410651 ] Committee Members: Sanjeev Arora(advisor), Jason D. Lee, Elad Hazan Abstract: In the past, traditional machine learning techniques relied on convex optimization for training. However, the deep learning revolution of the last decade has led to significant progress in many machine learning tasks, even though the loss landscape is highly non-convex. My research focuses on theoretical analysis of the training dynamics of deep neural nets, including convergence guarantees to low-loss solutions and implicit regularization effects of gradient-based methods. I will present our work on understanding the training dynamics in presence of normalization layers (e.g., Batch Normalization, Layer Normalization), which are important building blocks of modern neural nets. Our work shows the following: (1) normalization layers make the training process adaptive to different loss landscapes, even when using non-adaptive methods such as SGD; (2) normalization layers prevent full-batch gradient descent from converging to sharp minima and introduce an implicit bias towards flatter minima with better generalization, following a continuous sharpness-reduction flow. Reading List: [ https://docs.google.com/document/d/1BwImXphwwK5lk1NKDC0LApys0soGY7TVpptTZZz… | https://docs.google.com/document/d/1BwImXphwwK5lk1NKDC0LApys0soGY7TVpptTZZz… ] Everyone is invited to attend the talk, and those facultywishing to remain for the oral exam following are welcome to do so.

1 0

Re: [talks] UPDATED TIME Rachit Dubey will present his FPO "The successes and failures of human drives" on Friday, January 13, 2023 at 9:00 AM in CS 402.
by Grad Info 10 Jan '23

10 Jan '23

Please note, Rachit’s FPO start time has bene updated to 9:00 AM on Friday, January 13, 2023 in CS 402. Thank you. From: talks-bounces(a)lists.cs.princeton.edu <talks-bounces(a)lists.cs.princeton.edu> On Behalf Of CS Grad Department Sent: Tuesday, December 13, 2022 10:00 AM To: talks <talks(a)lists.cs.princeton.edu> Subject: [talks] Rachit Dubey will present his FPO "The successes and failures of human drives" on Friday, January 13, 2023 at 10:00 AM 9:00 AM in CS 402. Rachit Dubey will present his FPO "The successes and failures of human drives" on Friday, January 13, 2023 at 10:00 AM 9:00 AM in CS 402. Location: CS 402 The members of Rachit’s committee are as follows: Examiners: Tom Griffiths (Adviser), Ryan Adams, Jonathan Cohen Readers: Tania Lombrozo, Karthik Narasimhan A copy of his thesis is available upon request. Please email gradinfo(a)cs.princeton.edu <mailto:gradinfo@cs.princeton.edu> if you would like a copy of the thesis. Everyone is invited to attend his talk. Abstract follows below: Even in the absence of external rewards, we have internal motives that drive us to acquire information, pursue tasks, learn new things, etc. What is it that drives us? Under what conditions do these drives become maladaptive? In this dissertation, I employ computational modeling, behavioral experiments, and agent-based simulations to help develop a more complete picture of our intrinsic drives and motivations. In Chapter 2, I present a rational account of curiosity that unifies previous distinct theories in a single framework and explains a wide range of findings about human curiosity. Based on the insights from this framework, in Chapter 3, I present a behavioral intervention that can pique people’s curiosity for everyday scientific topics. Chapter 4 develops a computational model of Aha! moments and provides an explanation for why Aha! moments feel so rewarding. In Chapter 5, using the computational framework of reinforcement learning and the idea of reward design, I study the human drive to keep wanting more. I show that even though this seemingly maladaptive drive leads to unhappiness and overconsumption, it nevertheless plays an important role in promoting adaptive behavior and might be a deeply rooted bias of the human mind. Finally, in Chapter 6, I present an intervention that targets the wealthy and uses non-material social incentives to reduce their water consumption levels. Taken together, this work makes progress towards understanding the origins, strengths, and shortcomings of human drives, and illuminates the psychological forces that shape human behavior and suggests new ways to guide them.

1 0

Charlie Murphy will present his FPO "Relational Verification of Distributed Systems via Weak Simulations” on Friday, January 27, 2023 at 11am in CS 402.
by Nicki Mahler 06 Jan '23

06 Jan '23

Charlie Murphy will present his FPO "Relational Verification of Distributed Systems via Weak Simulations” on Friday, January 27, 2023 at 11am in CS 402. The members of his committee are as follows: Advisor: Zachary Kincaid Readers: Aarti Gupta and Lennart Beringer Examiners: Wyatt Lloyd and David Walker A copy of his thesis will be available before the FPO upon request. Please email gradinfo(a)cs.princeton.edu if you would like a copy of the thesis. Everyone is invited to attend his talk. Abstract: Distributed systems are notoriously hard to get right, even when based on existing protocols. This thesis explores the use of executable specifications for (semi-) automatic verification of distributed systems. Rather than proving a distributed system is correct with respect to a given temporal property, this thesis proposes instead verifying a distributed system implements a given distributed protocol via per-node simulation. To prove a distributed system implements a protocol this thesis defines (1) contextual simulation logic, (2) simulation synthesis, and (3) fine-grained strategy improvement. Contextual simulation logic is a relational Hoare logic that consists of a judgement called contextual simulation and a set of inference rules to prove a contextual simulation valid. A contextual simulation relates an implementation program to a specification program via divergence preserving weak simulation. The set of rules are defined using the structure of both programs and used to align the implementation and specification programs. Simulation synthesis is a semi-algorithm that may prove or refute the validity of a (per-node) contextual simulation. Simulation synthesis is based on the game semantics of contextual simulation. Simulation synthesis will (1) construct a complete well-labeled simulation unwinding (a finite representation of a winning strategy proving the validity of a contextual simulation), (2) a counter-strategy to simulation of finite duration (a winning strategy refuting the validity of a contextual simulation), or (3) diverge. To solve a simulation game---an infinite state game of infinite duration---simulation synthesis incrementally expands and solves a finite-horizon of the game. Finite-horizon games are solved by reduction to satisfiability of quantified linear integer arithmetic (LIA) formulas. Further, this thesis develops fine-grained strategy improvement a decision procedure for proving (or refuting) satisfiability of quantified LIA formulas based on the game-semantics of LIA formulas. This procedure produces a certificate (a winning strategy proving or refuting) satisfiability of a LIA Formula. This certificate is used to construct the (winning) strategy for the finite-horizon game.

1 0

Teague Tomesh will present his FPO entitled "Co-designing Quantum Computer Architectures and Algorithms to Bridge the Quantum Resource Gap" on Monday, January 23, 2023 at 11am in CS 105.
by Nicki Mahler 05 Jan '23

05 Jan '23

Teague Tomesh will present his FPO entitled "Co-designing Quantum Computer Architectures and Algorithms to Bridge the Quantum Resource Gap" on Monday, January 23, 2023 at 11am in CS 105. The members of Teague's committee are as follows: Examiners: Margaret Martonosi, Stephen Lyon, and Amit Levy Readers: Kyle Jamieson and Fred Chong (University of Chicago) A copy of his thesis will be available before the FPO upon request. Please email gradinfo(a)cs.princeton.edu if you would like a copy of the thesis. Everyone is invited to attend his talk. Abstract follows below: Quantum computing is a new computational paradigm based on the laws of quantum physics that have been developed over the last century. Quantum computers (QCs) manipulate quantum states and exploit non-classical phenomena, such as superposition and entanglement, to perform computations. Given this computational model, many quantum algorithms have been developed which are theoretically capable of outperforming any classical computer for certain applications such as factoring large integers, optimization, and simulating highly entangled quantum systems. However, the quantum programs implementing these high impact applications are extremely resource demanding. Their time and space requirements outstrip the capabilities of current QC systems by many orders of magnitude. I refer to this mismatch between the resources demanded by applications and what is available on current hardware as the Quantum Resource Gap (QRG). This dissertation presents a strategy for overcoming the QRG by advocating for the design of domain-specific quantum accelerators. I discuss how this strategy may be pursued using quantum benchmarks and program profiling to identify matches between applications and architectures that are well suited to one another. Once a particular application-architecture match is found, the algorithm’s execution can be optimized with cross-layer, co-design techniques that incorporate relevant information from across the entire hardware-software stack. To demonstrate the advantages of this approach, I discuss three examples covering molecular simulation, data set clustering, and constrained combinatorial optimization applications.

1 0

Paul Krueger will present his FPO "Metacognition: toward a computational framework for improving our minds" on Wednesday, January 18, 2023 at 2:00 PM in CS 402.
by gradinfo＠cs.princeton.edu 05 Jan '23

05 Jan '23

Paul Krueger will present his FPO "Metacognition: toward a computational framework for improving our minds" on Wednesday, January 18, 2023 at 2:00 PM in CS 402. Location: CS 402 The members of Paul's committee are as follows: Examiners: Tom Griffiths (Adviser), Jonathan Cohen, Ryan Adams Readers: Karthik Narasimhan, Nathaniel Daw A copy of his thesis will be available upon request. Please email gradinfo(a)cs.princeton.edu if you would like a copy of the thesis. Everyone is invited to attend his talk. Abstract follows below: In this dissertation I will show how reinforcement learning (RL) can be applied to the inner workings of cognition. The usual application of RL is to understand human behavior or build intelligent machines interacting in the external world. The same RL formalism can be inverted onto cognitive processes themselves, resulting in a normative account of how to explore and select mental computations, referred to as metacognitive RL. This framework can 1) be used to generate observable behavioral predictions, 2) provide a resource-rational benchmark for both assessing and improving cognition, and 3) motivate cognitive process models based on interacting RL systems. The formalism of metacognitive RL rests on meta-level Markov Decision Processes (meta-MDPs), which provide a general-purpose computational framework that can also make task-specific predictions. The first study applies the resource-rational framework to risky choice resulting in the identification of heuristics and accurate predictions about how people adapt their use of heuristics. The second study uses the same metacognitive RL framework to predict which structures of the environment will enhance metacognitive learning in humans during a planning task. In the third study, rather than manipulate the decision environment-which is often infeasible to do in the real-world-the metacognitive RL framework is used to produce feedback in the form of pseudorewards, resulting in faster metacognitive learning in a related planning task. Next, a new cooperative RL architecture is proposed. This approach also uses pseudorewards to promote learning, but rather than generate the pseudorewards from a computational model, it is proposed that they may be produced internally by a distinct RL system. The successful application of the metacognitive RL framework to understand and improve cognitive function depends critically on developing machine learning methods to solve these problems. In the final chapter, I briefly explore the application of a recently proposed machine learning method for solving meta-MDPs.

1 0

Paul Krueger will present his FPO "Metacognition: toward a computational framework for improving our minds" on Wednesday, January 18, 2023 at 2:00 PM in CS 402.
by Louis Riehl 22 Dec '22

22 Dec '22

Paul Krueger will present his FPO "Metacognition: toward a computational framework for improving our minds" on Wednesday, January 18, 2023 at 2:00 PM in CS 402. Location: CS 402 The members of Paul's committee are as follows: Examiners: Tom Griffiths (Adviser), Jonathan Cohen, Ryan Adams Readers: Karthik Narasimhan, Nathaniel Daw A copy of his thesis will be available two weeks before the FPO upon request. Please email gradinfo(a)cs.princeton.edu if you would like a copy of the thesis. Everyone is invited to attend his talk. Abstract follows below: Reinforcement learning has been used successfully to account for human decision-making and to produce intelligent behavior in artificial agents. However, people's internal cognitive processes generally remain obscure with these models. The same formalism used to account for external actions can also be applied to cognitive processes themselves. In particular, using the recently developed framework of resource rationality we can describe how people make efficient use of limited cognitive resources. Using this approach, we apply tools from machine learning to derive a normative account of how people ought to think efficiently. As a proof of concept, I will show how this framework can be used to automatically derive decision-making heuristics. I will present results from a classic multi-alternative risky choice task that allows us to trace people's cognitive processes. Using our method, we both re-discovered previously known heuristics, and discovered heuristics that had been overlooked. We found that people did indeed use the same heuristics that our method predicted, and furthermore adapted their heuristics to the structure of the decision environment in a rational way. Our approach bridges classic rationality with research on heuristics and biases by providing a general framework that can be used to apply rationality to any decision-making process while accounting for heuristics, thus providing a more realistic normative standard. In the second part of the talk, I will discuss a new framework for how model-based and model-free reinforcement learning systems can interact cooperatively, and the potential to account for human decision-making with this approach. In particular, this cooperation may be used to mediate the tension between goals and habits. I will close by discussing ongoing work to develop tools to solve meta-MDPs, the formalism used for our resource-rational approach.

1 0

Rachit Dubey will present his FPO "The successes and failures of human drives" on Friday, January 13, 2023 at 10:00 AM in CS 402.
by CS Grad Department 13 Dec '22

13 Dec '22

Rachit Dubey will present his FPO "The successes and failures of human drives" on Friday, January 13, 2023 at 10:00 AM in CS 402. Location: CS 402 The members of Rachit’s committee are as follows: Examiners: Tom Griffiths (Adviser), Ryan Adams, Jonathan Cohen Readers: Tania Lombrozo, Karthik Narasimhan A copy of his thesis is available upon request. Please email gradinfo(a)cs.princeton.edu if you would like a copy of the thesis. Everyone is invited to attend his talk. Abstract follows below: Even in the absence of external rewards, we have internal motives that drive us to acquire information, pursue tasks, learn new things, etc. What is it that drives us? Under what conditions do these drives become maladaptive? In this dissertation, I employ computational modeling, behavioral experiments, and agent-based simulations to help develop a more complete picture of our intrinsic drives and motivations. In Chapter 2, I present a rational account of curiosity that unifies previous distinct theories in a single framework and explains a wide range of findings about human curiosity. Based on the insights from this framework, in Chapter 3, I present a behavioral intervention that can pique people’s curiosity for everyday scientific topics. Chapter 4 develops a computational model of Aha! moments and provides an explanation for why Aha! moments feel so rewarding. In Chapter 5, using the computational framework of reinforcement learning and the idea of reward design, I study the human drive to keep wanting more. I show that even though this seemingly maladaptive drive leads to unhappiness and overconsumption, it nevertheless plays an important role in promoting adaptive behavior and might be a deeply rooted bias of the human mind. Finally, in Chapter 6, I present an intervention that targets the wealthy and uses non-material social incentives to reduce their water consumption levels. Taken together, this work makes progress towards understanding the origins, strengths, and shortcomings of human drives, and illuminates the psychological forces that shape human behavior and suggests new ways to guide them.

1 0

TODAY at 10am: Tyler Park Dissertation Defense (301 Computer Science Bldg)
by Nicki Mahler 07 Dec '22

07 Dec '22

The Lewis-Sigler Institute for Integrative Genomics ~ Graduate Program in Quantitative and Computational Biology Final Public Oral Tyler Park Integrating multi-modal biological data uncovers vulnerabilities and complex interactions in cancer Wednesday, December 7, 2022 10:00 AM Computer science building, room 301 Raphael Lab A copy of the dissertation is available to QCB for inspection electronically; email [ mailto:jgiraldi@princeton.edu | jgiraldi(a)princeton.edu ] with your request.

1 0

Yuxuan Zhang's MSE thesis talk "Defending Against Adversarial Attacks with Camera Image Pipelines" will take place December 6, 2022 at 5pm in CS 402
by Nicki Mahler 05 Dec '22

05 Dec '22

Yuxuan Zhang's MSE thesis talk "Defending Against Adversarial Attacks with Camera Image Pipelines" will take place December 6, 2022 at 5pm in CS 402. The members of his committee are Felix Heide (adviser) and Olga Russakovsky (reader). All are welcome to attend. Abstract: Existing neural networks for computer vision tasks are vulnerable to adversarial attacks: adding imperceptible perturbations to the input images can fool these models into making a false prediction on an image that was correctly predicted without the perturbation. Various defense methods have proposed image-to-image mapping methods, either including these perturbations in the training process or removing them in a preprocessing step. In doing so, existing methods often ignore that the natural RGB images in today’s datasets are not captured but, in fact, recovered from RAW color filter array captures that are subject to various degradations in the capture. In this work, we exploit this RAW data distribution as an empirical prior for adversarial defense. Specifically, we propose a model-agnostic adversarial defensive method, which maps the input RGB images to Bayer RAW space and back to output RGB using a learned camera image signal processing (ISP) pipeline to eliminate potential adversarial patterns. The proposed method acts as an off-the-shelf preprocessing module and, unlike model-specific adversarial training methods, does not require adversarial images to train. As a result, the method generalizes to unseen tasks without additional retraining. Experiments on large-scale datasets, e.g., ImageNet, COCO, for different vision tasks, e.g., classification, semantic segmentation, object detection, validate that the method significantly outperforms existing methods across task domains

1 0

Today! Eszter Hargittai Book Discussion- Connected in Isolation: Digital Privilege in Unsettled Times
by Emily C. Lawrence 30 Nov '22

30 Nov '22

[ https://citp.princeton.edu/event/hargittai-book-discussion/ | https://citp.princeton.edu/event/hargittai-book-discussion/ ] Wednesday, November 30, 2022 3 p.m. Bendheim House Join us for a discussion regarding [ https://nam12.safelinks.protection.outlook.com/?url=http%3A%2F%2Fbit.ly%2Fc… | Connected in Isolation: Digital Privilege in Unsettled Times ] by Eszter Hargittai, CITP Visiting Research Scholar. The book examines the hidden disparities that emerged during the global pandemic when access to education, work and even healthcare, hinged not only on access to the web, but certain digital skills, too. Hargittai will be joined by [ https://nam12.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.keithw… | Keith Wailoo ] , the Henry Putnam University Professor of History and Public Affairs at Princeton, CITP Executive Director [ https://citp.princeton.edu/citp-people/tithi-chattopadhyay/ | Tithi Chattopadhyay ] , and [ https://politics.princeton.edu/people/andrew-m-guess | Andy Guess ] , assistant professor of politics and public affairs at the School of Public and International Affairs to discuss the book. Bio: Eszter Hargittai is a professor and holds the Chair of Internet Use and Society at the Institute of Communication and Media Research of the University of Zurich. She is a fellow of the International Communication Association and an External Member of the Hungarian Academy of Sciences. She is past fellow of the Center for Advanced Study in the Behavioral Sciences at Stanford and Harvard’s Berkman Center for Internet & Society. She has also held visiting positions at Trinity College, Dublin (Ireland), the University of St. Gallen (Switzerland), University of Queensland (Australia), and the University of Vienna (Austria). Hargittai’s research looks at how people may benefit from or be left behind as a result of their varied digital media uses with a particular focus on how differences in people’s digital skills influence what they do online. She has looked at these questions in the domains of information seeking, health content (including Covid-19), political participation, job search, the sharing of creative content, and privacy management. Hargittai’s work has received awards from several professional associations including the International Communication Association’s Outstanding Young Scholar Award. For her teaching, she received the Galbut Outstanding Faculty Mentor Award of the School of Communication at Northwestern University, which honors one faculty member each year for exemplary teaching and engagement with students both inside and outside of the classroom. She has published over 100 journal articles, dozens of book chapters, and has edited [ http://eszter.com/books | four books ] . In addition to having presented her work across the US, she has also given invited talks in 15 countries on four continents. She has keynoted 18 meetings, given over 160 invited talks, and has presented at numerous conferences. Her work has been featured in many popular media outlets in the United States and internationally. Her research has been supported by the U.S. National Science Foundation, the John D. and Catherine T. MacArthur Foundation, the Alfred P. Sloan Foundation, Microsoft Research, Nokia, Google, Facebook, and Merck, among others. Hargittai is editor of the Handbook of Digital Inequality (Edward Elgar Publishing, 2021), Research Exposed: How Empirical Social Science Gets Done in the Digital Age (Columbia University Press 2021), Digital Research Confidential (with Christian Sandvig, The MIT Press, 2015), and Research Confidential: Solutions to Problems Most Social Scientists Pretend They Never Have (University of Michigan Press, 2009). The latter three present a behind-the-scenes look at doing empirical social science research. She has published academic career advice at Inside Higher Ed under the Ph.Do column. She has won some photo contests, has had her photography featured in books and calendars, and people she doesn’t know have been willing to part with their money to own some of her paintings. She tweets @eszter. If you cannot attend in person, you can [ https://princeton.zoom.us/j/96182136418 | click here to watch the webinar ] . Link for webinar: [ https://princeton.zoom.us/j/96182136418 | https://princeton.zoom.us/j/96182136418 ]

1 0