- talks - lists.cs.princeton.edu

Colloquium Speaker Christopher Manning, Thursday, Nov 16
by Emily Lawrence 08 Nov '17

08 Nov '17

Colloquium Speaker Prof. Christopher Manning, from Stanford University Thursday, November 16, 2017 - 12:30pm Computer Science - Room 105 Host: Dr. Sanjeev Arora *Please note, arrive early as seating and lunch may run out. Deep learning has had enormous success on perceptual tasks but still struggles in providing a model for inference. To address this gap, we have been developing Compositional Attention Networks (CANs). The CAN design provides a strong prior for explicitly iterative reasoning, enabling it to support explainable and structured learning, as well as generalization from a modest amount of data. The model builds on the great success of existing recurrent cells such as LSTMs: A CAN is a sequence of a single recurrent Memory, Attention, and Control (MAC) cell, and by careful design imposes structural constraints on the operation of each cell and the interactions between them, incorporating explicit control and soft attention mechanisms into their interfaces. We demonstrate the model's strength and robustness on the challenging CLEVR dataset for visual reasoning (Johnson et al. 2016), achieving a new state-of-the-art 98.9% accuracy, halving the error rate of the previous best model. More importantly, we show that the new model is more computationally efficient and data-efficient, requiring an order of magnitude less time and/or data to achieve good results. Joint work with Drew Arad. Bio: Christopher Manning is the Thomas M. Siebel Professor in Machine Learning, Linguistics and Computer Science at Stanford University. He works on software that can intelligently process, understand, and generate human language material. He is a leader in applying Deep Learning to Natural Language Processing, including exploring Tree Recursive Neural Networks, sentiment analysis, neural network dependency parsing, the GloVe model of word vectors, neural machine translation, and deep language understanding. He also focuses on computational linguistic approaches to parsing, robust textual inference and multilingual language processing, including being a principal developer of Stanford Dependencies and Universal Dependencies. Manning is an ACM Fellow, a AAAI Fellow, an ACL Fellow, and a Past President of ACL. He has coauthored leading textbooks on statistical natural language processing and information retrieval. He is a member of the Stanford NLP group (@stanfordnlp) and manages development of the Stanford CoreNLP software.

1 0

Distinguished Colloquium Speaker Lorrie Faith Cranor, Monday, Nov 13
by Emily Lawrence 07 Nov '17

07 Nov '17

Distinguished Colloquium Speaker Prof. Lorrie Faith Cranor, from Carnegie Mellon University Monday, November 13, 2017 - 12:30pm Computer Science - Room 105 Host: Dr. Marshini Chetty User studies are critical to understanding how users perceive and interact with security and privacy software and features. While it is important that users be able to configure and use security tools when they are not at risk, it is even more important that the tools continue to protect users during an attack. Conducting user studies in the presence of (simulated) risk is complicated. We would like to observe how users behave when they are actually at risk, but at the same time we cannot harm user study participants or subject them to increased risk. Often the risky situations we are interested in occur relatively infrequently in the real world, and thus can be difficult to observe in the wild. Researchers use a variety of strategies to overcome these challenges and place participants in situations where they will believe their security or privacy is at risk, without subjecting them to increases in actual harm. In some studies, researchers recruit participants to perform real tasks not directly related to security so that they can observe how participants respond to simulated security-related prompts or cues that occur while users are focused on primary tasks. In other studies, researchers create a hypothetical scenario and try to get participants sufficiently engaged in it that they will be motivated to avoid simulated harm. Sometimes researchers have the opportunity to observe real, rather than simulated attacks, although these opportunities are usually difficult to come by. Researchers can monitor real world user behavior over long periods of time (in public or with permission of participants) and observe how users respond to risks that occur naturally, without researcher intervention. In this talk I will motivate the importance of security user studies and talk about a number of different user study approaches we have used at the CyLab Usable Privacy and Security Lab at Carnegie Mellon University. Lorrie Faith Cranor is a Professor of Computer Science and of Engineering and Public Policy at Carnegie Mellon University where she is director of the CyLab Usable Privacy and Security Laboratory (CUPS). She is associate department head of the Engineering and Public Policy Department and co-director of the MSIT-Privacy Engineering masters program. In 2016 she served as Chief Technologist at the US Federal Trade Commission, working in the office of Chairwoman Ramirez. She is also a co-founder of Wombat Security Technologies, Inc, a security awareness training company. She has authored over 150 research papers on online privacy, usable security, and other topics. She has played a key role in building the usable privacy and security research community, having co-edited the seminal book Security and Usability (O'Reilly 2005) and founded the Symposium On Usable Privacy and Security (SOUPS). She also chaired the Platform for Privacy Preferences Project (P3P) Specification Working Group at the W3C and authored the book Web Privacy with P3P (O'Reilly 2002). She has served on a number of boards, including the Electronic Frontier Foundation Board of Directors, and on the editorial boards of several journals. In her younger days she was honored as one of the top 100 innovators 35 or younger by Technology Review magazine. More recently she was elected to the ACM CHI Academy, named an ACM Fellow for her contributions to usable privacy and security research and education, and named an IEEE Fellow for her contributions to privacy engineering. She was previously a researcher at AT&T-Labs Research and taught in the Stern School of Business at New York University. She holds a doctorate in Engineering and Policy from Washington University in St. Louis. In 2012-13 she spent her sabbatical as a fellow in the Frank-Ratchye STUDIO for Creative Inquiry at Carnegie Mellon University where she worked on fiber arts projects that combined her interests in privacy and security, quilting, computers, and technology. She practices yoga, plays soccer, and runs after her three children.

1 0

Distinguished Colloquium Speaker Jeff Heer, Today!
by Jennifer Cirillo 06 Nov '17

06 Nov '17

Distinguished Colloquium Speaker Professor Jeffrey Michael Heer, University of Washington Monday, November 6, 2017 - 12:30pm Computer Science - Room 105 Host: Adam Finkelstein Data analysis is a complex process with frequent shifts among data formats, tools and models, as well as between symbolic and visual thinking. How might the design of improved tools accelerate people's exploration and understanding of data? Covering both interactive demos and principles from academic research, this talk will examine how to craft a careful balance of interactive and automated methods, combining concepts from data visualization, machine learning, and computer systems to design novel interactive analysis tools. Jeffrey Heer is an Associate Professor of Computer Science & Engineering at the University of Washington, where he directs the Interactive Data Lab and conducts research on data visualization, human-computer interaction and social computing. The visualization tools developed by Jeff and his collaborators (Vega, D3.js, Protovis, Prefuse) are used by researchers, companies, and thousands of data enthusiasts around the world. Jeff’s research papers have received awards at the premier venues in Human-Computer Interaction and Visualization (ACM CHI, UIST, CSCW; IEEE InfoVis, VAST, EuroVis). Other honors include MIT Technology Review’s TR35 (2009), a Sloan Fellowship (2012), and the ACM Grace Murray Hopper Award (2016). Jeff holds B.S., M.S., and Ph.D. degrees in Computer Science from UC Berkeley, whom he then betrayed to join the Stanford faculty (2009–2013). He is also a co-founder of Trifacta, a provider of interactive tools for scalable data transformation.

1 0

Fwd: CS/EE Seminar - Monday, November 6. 2017, 4:30 pm, Computer Science 105 - Prof. Benjamin Lee
by Margaret Martonosi 05 Nov '17

05 Nov '17

FYI: Prof. Ben Lee from Duke University will give a seminar tomorrow, Monday Nov 6 at 4:30 in the CS Small Auditorium on "Rethinking Datacenter Management with Game Theory”. Full info below. -mrm Margaret Martonosi H. T. Adams '35 Professor of Computer Science. Director, Keller Center for Innovation in Engineering Education Princeton University ■ mrm(a)princeton.edu ■ http://www.princeton.edu/~mrm > Begin forwarded message: > > From: "Lori A. Bailey" <lbailey(a)Princeton.EDU> > Subject: CS/EE Seminar - Monday, November 6. 2017, 4:30 pm, Computer Science 105 - Prof. Benjamin Lee > Date: October 23, 2017 at 11:59:27 AM EDT > To: ee-seminar(a)Princeton.EDU > Reply-To: "Lori A. Bailey" <lbailey(a)Princeton.EDU> > > > > > Department of Computer Science and > Department of Electrical Engineering Seminar > > > > Speaker: Prof. Benjamin Lee > Duke University > > Title: Rethinking Datacenter Management with Game Theory > Date: Monday, November 6, 2017 > Time: 4:30 p.m. > Room: Computer Science, Room 105 > Host: Prof. Margaret Martonosi and Prof. David Wentzlaff > > Abstract: Sharing datacenter hardware improves energy efficiency, but whether strategic users participate in consolidated systems depends on management policies. Users who dislike allocations may refuse to participate and deploy private, less-efficient systems. We rethink systems management, drawing on game theory to model strategic behavior and incentivize participation. We illustrate this perspective for three fundamental challenges in datacenters. For workload colocation, we design matching games to produce stable coalitions in which more users share with preferred co-runners. For resource allocation, we use Cobb-Douglas utility functions to produce fair allocations that incentivize users to share cache and memory. For power delivery, we design sprinting games to produce equilibria in which users selfishly draw power for performance boosts yet avoid oversubscribing the shared supply. Collectively, these varied solution concepts provide a rich toolkit for rigorously managing datacenters shared by strategic and competitive users. > > Biography: Benjamin Lee is an Associate Professor of Electrical and Computer Engineering at Duke University. His research focuses on computer architectures, security / privacy, and energy efficiency with links to statistical inference and algorithmic economics. His research has been recognized by IEEE Micro Top Picks (4x), Communications of the ACM Research Highlights (2x), as well as varied paper honors from ASPLOS, MICRO, and SC. He received the NSF Computing Innovation Fellowship, NSF CAREER Award, and Google Faculty Research Award. He held visiting positions at Microsoft Research, Intel Labs, and Lawrence Livermore National Lab. Dr. Lee received his B.S. from the University of California at Berkeley, his Ph.D. from Harvard University, and his post-doctorate from Stanford University.

1 0

Distinguished Colloquium Speaker Jeff Heer, Monday, November 6th - 12:30pm
by Jennifer Cirillo 05 Nov '17

05 Nov '17

Distinguished Colloquium Speaker Professor Jeffrey Michael Heer, University of Washington Monday, November 6, 2017 - 12:30pm Computer Science - Room 105 Host: Adam Finkelstein Data analysis is a complex process with frequent shifts among data formats, tools and models, as well as between symbolic and visual thinking. How might the design of improved tools accelerate people's exploration and understanding of data? Covering both interactive demos and principles from academic research, this talk will examine how to craft a careful balance of interactive and automated methods, combining concepts from data visualization, machine learning, and computer systems to design novel interactive analysis tools. Jeffrey Heer is an Associate Professor of Computer Science & Engineering at the University of Washington, where he directs the Interactive Data Lab and conducts research on data visualization, human-computer interaction and social computing. The visualization tools developed by Jeff and his collaborators (Vega, D3.js, Protovis, Prefuse) are used by researchers, companies, and thousands of data enthusiasts around the world. Jeff’s research papers have received awards at the premier venues in Human-Computer Interaction and Visualization (ACM CHI, UIST, CSCW; IEEE InfoVis, VAST, EuroVis). Other honors include MIT Technology Review’s TR35 (2009), a Sloan Fellowship (2012), and the ACM Grace Murray Hopper Award (2016). Jeff holds B.S., M.S., and Ph.D. degrees in Computer Science from UC Berkeley, whom he then betrayed to join the Stanford faculty (2009–2013). He is also a co-founder of Trifacta, a provider of interactive tools for scalable data transformation.

1 0

Distinguished Colloquium Speaker Jeff Heer, Monday, November 6th - 12:30pm
by Jennifer Cirillo 31 Oct '17

31 Oct '17

Distinguished Colloquium Speaker Professor Jeffrey Michael Heer, University of Washington Monday, November 6, 2017 - 12:30pm Computer Science - Room 105 Host: Adam Finkelstein Data analysis is a complex process with frequent shifts among data formats, tools and models, as well as between symbolic and visual thinking. How might the design of improved tools accelerate people's exploration and understanding of data? Covering both interactive demos and principles from academic research, this talk will examine how to craft a careful balance of interactive and automated methods, combining concepts from data visualization, machine learning, and computer systems to design novel interactive analysis tools. Jeffrey Heer is an Associate Professor of Computer Science & Engineering at the University of Washington, where he directs the Interactive Data Lab and conducts research on data visualization, human-computer interaction and social computing. The visualization tools developed by Jeff and his collaborators (Vega, D3.js, Protovis, Prefuse) are used by researchers, companies, and thousands of data enthusiasts around the world. Jeff’s research papers have received awards at the premier venues in Human-Computer Interaction and Visualization (ACM CHI, UIST, CSCW; IEEE InfoVis, VAST, EuroVis). Other honors include MIT Technology Review’s TR35 (2009), a Sloan Fellowship (2012), and the ACM Grace Murray Hopper Award (2016). Jeff holds B.S., M.S., and Ph.D. degrees in Computer Science from UC Berkeley, whom he then betrayed to join the Stanford faculty (2009–2013). He is also a co-founder of Trifacta, a provider of interactive tools for scalable data transformation.

1 0

popular talk on ML at IAS today 5:30pm
by Sanjeev Arora 27 Oct '17

27 Oct '17

I am involved in a popular lecture + panel discussion today on ML at IAS. 5:30pm. https://www.ias.edu/events/arora-zemel-publiclecture Best Sanjeev

1 0

Yonatan Naamad will present his FPO, "Hardness from Densest Subgraph Conjectures"on Monday, October 30, 2017 at 4pm in CS 402
by Nicki Gotsis 24 Oct '17

24 Oct '17

Yonatan Naamad will present his FPO, "Hardness from Densest Subgraph Conjectures"on Monday, October 30, 2017 at 4pm in CS 402. The members of his committee are: Moses Charikar (adviser); Readers: Sanjeev Arora and Anthony Wirth (University of Melbourne); Examiners: Moses Charikar, Mark Braverman, and Matt Weinberg. A copy of his thesis is available in Room 310. Everyone is invited to attend his talk. The talk abstract is attached to this email.

1 1

Fwd: CS/EE Seminar - Monday, November 6. 2017, 4:30 pm, Computer Science 105 - Prof. Benjamin Lee
by Jennifer Rexford 23 Oct '17

23 Oct '17

> > From: "Lori A. Bailey" <lbailey(a)Princeton.EDU> > Subject: CS/EE Seminar - Monday, November 6. 2017, 4:30 pm, Computer Science 105 - Prof. Benjamin Lee > Date: October 23, 2017 at 11:59:27 AM EDT > To: ee-seminar(a)Princeton.EDU > Reply-To: "Lori A. Bailey" <lbailey(a)Princeton.EDU> > > > > > Department of Computer Science and > Department of Electrical Engineering Seminar > > > > Speaker: Prof. Benjamin Lee > Duke University > > Title: Rethinking Datacenter Management with Game Theory > Date: Monday, November 6, 2017 > Time: 4:30 p.m. > Room: Computer Science, Room 105 > Host: Prof. Margaret Martonosi and Prof. David Wentzlaff > > Abstract: Sharing datacenter hardware improves energy efficiency, but whether strategic users participate in consolidated systems depends on management policies. Users who dislike allocations may refuse to participate and deploy private, less-efficient systems. We rethink systems management, drawing on game theory to model strategic behavior and incentivize participation. We illustrate this perspective for three fundamental challenges in datacenters. For workload colocation, we design matching games to produce stable coalitions in which more users share with preferred co-runners. For resource allocation, we use Cobb-Douglas utility functions to produce fair allocations that incentivize users to share cache and memory. For power delivery, we design sprinting games to produce equilibria in which users selfishly draw power for performance boosts yet avoid oversubscribing the shared supply. Collectively, these varied solution concepts provide a rich toolkit for rigorously managing datacenters shared by strategic and competitive users. > > Biography: Benjamin Lee is an Associate Professor of Electrical and Computer Engineering at Duke University. His research focuses on computer architectures, security / privacy, and energy efficiency with links to statistical inference and algorithmic economics. His research has been recognized by IEEE Micro Top Picks (4x), Communications of the ACM Research Highlights (2x), as well as varied paper honors from ASPLOS, MICRO, and SC. He received the NSF Computing Innovation Fellowship, NSF CAREER Award, and Google Faculty Research Award. He held visiting positions at Microsoft Research, Intel Labs, and Lawrence Livermore National Lab. Dr. Lee received his B.S. from the University of California at Berkeley, his Ph.D. from Harvard University, and his post-doctorate from Stanford University.

1 0

Andrej Risteski will present his FPO, "New techniques for learning and inference in Bayesian models" on Thursday, October 26, 2017 at 1pm in CS 402
by Nicki Gotsis 19 Oct '17

19 Oct '17

Andrej Risteski will present his FPO, "New techniques for learning and inference in Bayesian models" on Thursday, October 26, 2017 at 1pm in CS 402. The members of his committee are: Readers: Sanjeev Arora (adviser), Rong Ge (Duke University), and Elad Hazan Nonreaders: Sanjeev Arora, Elad Hazan, Bernard Chazelle, and Rong Ge (Duke University), and Mark Braverman A copy of his thesis, is available in Room 310. Everyone is invited to attend his talk. The talk abstract follows below. Abstract: A common theme in machine learning is succinct modeling of distributions over large domains. Probabilistic graphical models are one of the most expressive frameworks for doing this. The two major tasks involving graphical models are learning and inference. Learning is the task of calculating the best fit graphical model from raw data, while inference is the task of answering probabilistic queries for a known graphical model (for example what is the marginal distribution of one of the variables or what is the distribution of a subset of variables, after conditioning on the values of some other subset of variables). Learning can be thought of as finding a graphical model that explains the raw data, while the inference queries extract the knowledge the graphical model contains. This thesis introduces new provable techniques for performing both of these tasks, in the context of both latent-variable models – in which a portion of the variables in the graphical model are not observed, as well as fully-observable undirected graphical models (Markov Random Fields). Chapters 2 and 3 will focus on learning latent-variable models, while Chapter 4 will focus on inference in Markov Random Fields. In Chapter 2, I will contribute the first provable results for analyzing variational Bayes: a family of alternatingminimization style algorithms which is very popular in practice for learning latent-variable models. Despite it’s popularity with practitioners, the only theoretical guarantees prior to this work concerned convergence to local minima. We will prove that under reasonable assumptions, in the context of topic models, these algorithms will converge to the global minimum. Subsequently, in Chapter 3, we will use the method-of-moments along with new techniques in tensor decomposition and constrained matrix factorization to derive algorithms for learning noisy-OR networks – the textbook example of a probabilistic model for causal relationships. Importantly, these techniques were only applicable to linear latentvariable models – which noisy-OR is not. In Chapter 4, I will contribute a new understanding of a class of variational methods for calculating partition functions in Markov Random Fields. The key technical ingredient is a connection to convex programming hierarchies – a recent area of interest in combinatorial optimization, along with approximations of the entropy of a distribution based on low-order moment information.

1 0