- talks - lists.cs.princeton.edu

Donghun Lee will present his Pre-FPO, "On Finite Horizon Behaviors of Machine Learning: the Max-Operator Bias in Q-learning and the Framework of Learning to Learn Optimally" on Thursday, November 29, 2018 at 1:30pm in CS 105
by Nicki Gotsis 27 Nov '18

27 Nov '18

Donghun Lee will present his Pre-FPO on Thursday, November 29, 2018 at 1:30pm in CS 105. The members of his committee are Warren Powell (ORFE) (adviser), Ryan Adams, and Peter Ramadge (EE). Everyone is invited to attend his talk. The talk title and abstract follow below. Title: On Finite Horizon Behaviors of Machine Learning: the Max-Operator Bias in Q-learning and the Framework of Learning to Learn Optimally Abstract: With greater popularity of machine learning in industry and growing number of available packages and toolkits, many end users of machine learning face the problem of deploying machine learning algorithms to their applications. Most algorithms with asymptotic guarantees leave finite-time issues such as initialization or tuning at the mercy of the end users, to whom the burden may cause undesirable outcomes in practice with finite time horizon. This talk will cover my research on finite time horizon behavior of machine learning algorithms: firstly on Q-learning algorithm, and secondly on the problem of "learning to learn optimally". First, I will review the positive bias innate in Q-learning algorithm, and present a method to correct the bias on-the-fly. Then, I will address the problem of applying machine learning algorithm to finite time horizon problems to formulate the "learning to learn optimally" framework. Using the framework, I will construct a knowledge-gradient based algorithm for adaptive deployment of machine learning application with finite time horizon. Some work in progress will be summarized at the end of the talk.

1 0

Fwd: Thursday, November 29, 4:30pm, E-Quad B205 - Pramod Viswanath, University of Illinois at Urbana-Champaign
by Jennifer Rexford 26 Nov '18

26 Nov '18

> > From: EE Executive Office <eemain(a)Princeton.EDU> > Subject: Thursday, November 29, 4:30pm, E-Quad B205 - Pramod Viswanath, University of Illinois at Urbana-Champaign > Date: November 26, 2018 at 10:38:23 AM EST > To: ee-seminar(a)Princeton.EDU > Reply-To: EE Executive Office <eemain(a)Princeton.EDU> > > > > EE SEMINAR SERIES > > > Speaker: > Pramod Viswanath, University of Illinois at Urbana-Champaign > Title: > Why Blocks and Why Chains; A First Principles (Re)Design of Blockchains > Day: > Thursday, November 29, 2018 > Time: > 4:30 pm > Room: > B205 Engineering Quadrangle > Host: > Prof. Naveen Verma > > > Abstract: > Today's blockchains do not scale in a meaningful way. As more nodes join the system, the efficiency of the system (computation, communication, and storage) degrades, or at best stays constant. Furthermore, the security of the permission less system imposes limitations on the core performance metrics of throughput, latency and reliability. We take a first principle approach to the blockchain ecosystem addressing each of the various components holistically. Our approach is characterized by seeking fundamental limits (those prescribed by the physics of the underlying network) to performance and designing algorithms that attain them. This research is informed by decades of experience in information theory, coding theory, algorithms, wireless communication and packet networks. This talk will highlight key outcomes of this research program, including: > Prism (a new consensus algorithm that guarantees information theoretically optimal throughput, latency, reliability), > Spider (a new networking protocol for off-chain payment channels), > Polyshard (a new coded storage architecture), and > Dandelion (a new network privacy layer). > > References: > https://arxiv.org/abs/1810.08092 <https://arxiv.org/abs/1810.08092> > https://arxiv.org/abs/1809.10361 <https://arxiv.org/abs/1809.10361> > https://arxiv.org/abs/1809.05088 <https://arxiv.org/abs/1809.05088> > https://arxiv.org/abs/1809.07468 <https://arxiv.org/abs/1809.07468> > > > Bio: > Pramod Viswanath received the Ph.D. degree in EECS from UC Berkeley in 2000. From 2000 to 2001, he was a member of research staff at Flarion technologies, NJ. Since 2001, he is on the faculty at University of Illinois at Urbana Champaign in Electrical and Computer Engineering, where he currently is a professor. He is a coauthor, with David Tse, of the text Fundamentals of Wireless Communication, which has been used in over 60 institutions around the world. He is coinventor of the opportunistic beamforming method and codesigner of Flash-OFDM communication algorithms adapted into fourth-generation cellular systems. > > His current research interests are in blockchain technologies from a variety of angles: networking protocols, consensus algorithms, payment channels, distributed coded storage and incentive designs. He is co-founder and CEO of Applied Protocol Research, a startup doing research on blockchain technologies. Applied Protocol Research is staffed by academics (professors, PhDs, and intern graduate students), with a wide variety of backgrounds (EE/CS/ECON covering both theory/systems from different institutions (Berkeley, CMU, Illinois, MIT, Stanford, USC, UW-Seattle). > > This talk is joint work by the speaker with: Mohammad Alizadeh (MIT), Salman Avestimehr (USC), Giulia Fanti (CMU), Sreeram Kannan (UW-Seattle), Sewoong Oh (Illinois) and David Tse (Stanford).

1 0

CANCELED: Colloquium Speaker-Andy Pavlo
by Emily C. Lawrence 16 Nov '18

16 Nov '18

Due to yesterday's weather events, today's talk has been canceled. Colloquium Speaker Andy Pavlo from Carnegie Mellon University Friday, November 16 - 12:30pm Computer Science - Room 105 “Make Your Database Dream of Electric Sheep: Designing for Autonomous Operation” In the last 20 years, researchers and vendors have built advisory tools to assist DBAs in tuning and physical design. Most of this previous work is incomplete because they require humans to make the final decisions about any database changes and are reactionary measures that fix problems after they occur. What is needed for a "self-driving" DBMS are components that are designed for autonomous operation. This will enable new optimizations that are not possible today because the complexity of managing these systems has surpassed the abilities of humans. In this talk, I present the core design principles of an autonomous DBMS. These are necessary to support ample data collection, fast state changes, and accurate reward observations. I will discuss techniques on how to build a new autonomous DBMS or the steps needed to retrofit an existing one to enable automated management. Our work is based on our experiences at CMU from developing an automatic tuning service ( OtterTune ) and our self-driving DBMS (Peloton). Bio: Andy Pavlo is an Assistant Professor of Databaseology in the Computer Science Department at Carnegie Mellon University. He also used to raise clams.

1 0

REMINDER: Colloquium Speaker-Andy Pavlo, tomorrow at 12:30pm
by Emily C. Lawrence 15 Nov '18

15 Nov '18

Colloquium Speaker Andy Pavlo from Carnegie Mellon University Friday, November 16 - 12:30pm Computer Science - Room 105 “Make Your Database Dream of Electric Sheep: Designing for Autonomous Operation” In the last 20 years, researchers and vendors have built advisory tools to assist DBAs in tuning and physical design. Most of this previous work is incomplete because they require humans to make the final decisions about any database changes and are reactionary measures that fix problems after they occur. What is needed for a "self-driving" DBMS are components that are designed for autonomous operation. This will enable new optimizations that are not possible today because the complexity of managing these systems has surpassed the abilities of humans. In this talk, I present the core design principles of an autonomous DBMS. These are necessary to support ample data collection, fast state changes, and accurate reward observations. I will discuss techniques on how to build a new autonomous DBMS or the steps needed to retrofit an existing one to enable automated management. Our work is based on our experiences at CMU from developing an automatic tuning service ( OtterTune ) and our self-driving DBMS (Peloton). Bio: Andy Pavlo is an Assistant Professor of Databaseology in the Computer Science Department at Carnegie Mellon University. He also used to raise clams.

1 0

Nora Willett will present her Pre FPO on Tuesday Nov, 27th, 2018 at 3pm in CS 402.
by Nicki Gotsis 13 Nov '18

13 Nov '18

Nora Willett will present her Pre FPO on Tuesday Nov, 27th, 2018 at 3pm in CS 402. The members of her committee are: Adam Finkelstein (Adviser), Szymon Rusinkiewicz, Marshini Chetty, Tom Funkhouser, and Wilmot Li (Adobe). Everyone is invited to attend her talk. The talk title and abstract follow below: Title : Animation Authoring Tools Abstract : Live animation of 2D characters is an emerging form of storytelling that has begun to appear on streaming platforms and broadcast TV. Unlike traditional animation, human performers control characters in real time so that they can respond and improvise to live events. Current live animation systems provide a range of animation controls, such as camera input to drive head movements, audio for lip sync, and keyboard shortcuts to trigger discrete pose changes via artwork swaps. However, managing all of these controls during a live performance is challenging. In this talk, I present three tools to assist performers in creating compelling live 2D animation. First, I will introduce a new interactive system that addresses the problem of triggering artwork swaps in live settings. Next, I will describe a framework that augments the primary motion of a figure by adding secondary motion – subtle movement of parts like hair, foliage or cloth that complements and emphasizes the primary motion. Finally, I will show a mixed initiative interface which, from a few user-drawn scribbles, extracts repetitive texture elements in an image, and supports animating them.

1 0

Tomorrow: Talk by Armando Solar-Lezama, Wednesday, Nov 14
by Emily Lawrence 13 Nov '18

13 Nov '18

PL Seminar Speaker Armando Solar-Lezama, from MIT Wednesday, November 14 - 3:00 pm Computer Science - Room 302 Formal Methods meets ML: Synthesis, Verification and Beyond In this talk I will describe some recent work in my group that brings together machine learning and formal methods. The talk will describe both how formal methods can feed off from each other to the benefit of both. Specifically, I will highlight three recent projects where the interaction of the two leads to new capabilities for both. First, I will talk about how ideas from AI and Machine learning can benefit program synthesis and lead to synthesis systems that are more flexible and easier to develop. I will also describe how ideas from formal methods can be used both to verify machine learning systems and to gain a better understanding of their behavior. Bio: Armando Solar-Lezama is an associate professor in the department of electrical engineering and computer science at MIT. He leads the Computer Aided Programming group, which has been working on the application of learning and automated reasoning to automate challenging aspects of programming.

1 0

Talk by Byron Cook, Friday, Nov 16
by Emily Lawrence 13 Nov '18

13 Nov '18

PL Seminar Speaker Byron Cook, from Amazon Web Services and UCL Friday, November 16 - 2:00 pm Computer Science - Room 302 Reasoning about Security of Amazon Web Services This talk will discuss the development and use of formal verification tools within Amazon Web Services (AWS) to increase the security assurance of its cloud infrastructure and to help customers secure themselves. I’ll also discuss some remaining challenges that could inspire future research in the community. Bio: Byron Cook is Professor of Computer Science at University College London (UCL) and Director of Automated Reasoning at Amazon Web Services. Byron’s interests include computer/network security, program analysis/verification, programming languages, theorem proving, logic, hardware design, operating systems, and biological systems. Byron is the founder and leader of Amazon’s Automated Reasoning Group (ARG).

1 0

TODAY November 12, 4:30pm, E-Quad B205 - Young-Han Kim, University of California, San Diego
by Emily Lawrence 12 Nov '18

12 Nov '18

EE SEMINAR SERIES Speaker: Young-Han Kim, University of California, San Diego Title: Data Science with Universal Probability Day: Monday, November 12, 2018 Time: 4:30 pm Room: B205 Engineering Quadrangle Host: Prof. Yuxin Chen Abstract: As Laplace famously asked "What is the probability that the sun will rise tomorrow?," inferring the probabilities underlying a given data is at the heart of many data science problems. In this talk, I will explore how to assign probabilities to data such that its unknown distribution is uniformly approximated without overfitting the data. I will develop the mathematical foundation of this theory of universal probability which traces back to Rissanen, to Ziv and Lempel, and even to Laplace, and present a general framework for addressing many data science problems in a unified manner. To illustrate this framework, I will present two practical applications in nucleotide sequence classification and grayscale image denoising. Bio: Young-Han Kim received his B.S. degree in Electrical Engineering from Seoul National University in 1996 and his Ph.D. degree in Electrical Engineering (M.S. degrees in Statistics and in Electrical Engineering) from Stanford University in 2006. Since then he has been on the faculty of the Department of Electrical and Computer Engineering at the University of California, San Diego, where he is currently a Professor. His research contributions have been in information theory, communication engineering, and data science. He has coauthored a highly cited textbook ``Network Information Theory'' (Cambridge University Press, 2011) and a recent monograph ``Fundamentals of Index Coding'' (Now Publishers, 2018). He has received several awards and honors for his contributions, including the NSF CAREER Award (2008), the US-Israel BSF Bergmann Memorial Award (2009), the IEEE Information Theory Paper Award (2012), and the first IEEE James L. Massey Research and Teaching Award (2015). He is a fellow of the IEEE. This seminar is supported with funds from the Korhammer Lecture Series

1 0

Colloquium Speaker-Andy Pavlo, Friday, Nov 16
by Emily C. Lawrence 12 Nov '18

12 Nov '18

Colloquium Speaker Andy Pavlo from Carnegie Mellon University Friday, November 16 - 12:30pm Computer Science - Room 105 “Make Your Database Dream of Electric Sheep: Designing for Autonomous Operation” In the last 20 years, researchers and vendors have built advisory tools to assist DBAs in tuning and physical design. Most of this previous work is incomplete because they require humans to make the final decisions about any database changes and are reactionary measures that fix problems after they occur. What is needed for a "self-driving" DBMS are components that are designed for autonomous operation. This will enable new optimizations that are not possible today because the complexity of managing these systems has surpassed the abilities of humans. In this talk, I present the core design principles of an autonomous DBMS. These are necessary to support ample data collection, fast state changes, and accurate reward observations. I will discuss techniques on how to build a new autonomous DBMS or the steps needed to retrofit an existing one to enable automated management. Our work is based on our experiences at CMU from developing an automatic tuning service ( OtterTune ) and our self-driving DBMS (Peloton). Bio: Andy Pavlo is an Assistant Professor of Databaseology in the Computer Science Department at Carnegie Mellon University. He also used to raise clams.

1 0

Wednesday, November 14, 12:30pm, E-Quad B205 - Gongguo Tang, Colorado School of Mines
by Emily Lawrence 09 Nov '18

09 Nov '18

EE SEMINAR SERIES Speaker: Gongguo Tang, Colorado School of Mines Title: Optimal Spectral Estimation via Atomic Norm Minimization Day: Wednesday, November 14, 2018 Time: 12:30 pm Room: B205 Engineering Quadrangle Host: Prof. Yuxin Chen Abstract: Atomic norm minimization is a convex relaxation framework that greatly generalizes l1 norm for compressed sensing and nuclear norm for matrix completion. In particular, it allows one to construct convex regularizers for signals that have sparse representations with respect to continuously parameterized dictionaries. In this talk, the speaker will focus on the application of this framework to line spectral estimation, which can be viewed as a sparse recovery problem whose atoms are indexed by the continuous frequency variable. In particular, the method's accuracy in inferring the frequencies and complex magnitudes from noisy observations of a mixture of complex sinusoids will be analyzed. When the Signal-to-Noise Ratio is reasonably high and the true frequencies are well-separated, the atomic norm estimator is shown to localize the correct number of frequencies, each within a neighborhood of one of the true frequencies, whose size matches the CramérRao lower bound up to a logarithmic factor. The analysis is based on a primaldual witness construction procedure. The analysis also reveals that the atomic norm minimization can be viewed as a convex way to solve a l1-norm regularized, nonlinear and nonconvex least-squares problem to global optimality. Bio: Dr. Gongguo Tang is an Assistant Professor in the Department of Electrical Engineering at Colorado School of Mines since 2014. He received his Ph.D. degree in Electrical Engineering from Washington University in St. Louis in 2011. He was a Postdoctoral Research Associate at the Department of Electrical and Computer Engineering, University of Wisconsin-Madison from 2011 to 2013, and a visiting scholar at the University of California, Berkeley in 2013. Dr. Tang's research interests are in the area of optimization, signal processing, and machine learning, and their applications in big data analytics, optics, imaging, and networks. This seminar is supported with funds from the Korhammer Lecture Series

1 0