BEGIN:VCALENDAR
METHOD:PUBLISH
PRODID:-//Apple Computer\, Inc//iCal 1.0//EN
X-WR-CALNAME;VALUE=TEXT:USC
VERSION:2.0
BEGIN:VEVENT
DESCRIPTION:Speaker: Varun Jog, Assistant Professor, University of Wisconsin-Madison
Talk Title: Analyzing Learning Algorithms: Perspectives from Information Theory and Optimal Transport
Abstract: In this talk, we will analyze generalization and robustness properties of learning algorithms using tools derived from information theory and optimal transport. In statistical learning theory, generalization error is used to quantify the degree to which a supervised machine learning algorithm may overfit to training data. Leveraging recent work [Xu and Raginsky (2017)], we derive information-theoretic generalization error bounds for a broad class of iterative algorithms that are characterized by bounded, noisy updates with Markovian structure, such as stochastic gradient Langevin dynamics (SGLD). We describe certain shortcomings of these information-theoretic bounds, and propose alternate strategies that rely on optimal transport theory. We show that results from optimal transport are well-suited to analyze not only generalization properties, but also robustness properties of learning algorithms.
Biography: Varun Jog received his B.Tech. degree in Electrical Engineering from IIT Bombay in 2010, and his Ph.D. in Electrical Engineering and Computer Sciences (EECS) from UC Berkeley in 2015. Since 2016, he is an Assistant Professor at the Electrical and Computer Engineering Department and a fellow at the Grainger Institute for Engineering at the University of Wisconsin - Madison. His research interests include information theory, machine learning, and network science. He is a recipient of the Eli Jury award from the EECS Department at UC Berkeley (2015) and the Jack Keil Wolf student paper award at ISIT 2015.\n
Host: Professor Salman Avestimehr, avestime@usc.edu
SEQUENCE:5
DTSTART:20190422T110000
LOCATION:EEB 248
DTSTAMP:20190422T110000
SUMMARY:ECE Seminar: Analyzing Learning Algorithms: Perspectives from Information Theory and Optimal Transport
UID:EC9439B1-FF65-11D6-9973-003065F99D04
DTEND:20190422T120000
END:VEVENT
END:VCALENDAR