BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Seminar by Prof. Huan Xu\, Georgia Tech
DTSTART:20191217T143000
DTEND:20191217T160000
DTSTAMP:20260929T050204Z
UID:4e7e6941ff44240a7782400cf5236c65730a6f0e36af0c4e6d15b99a
CATEGORIES:Conferences - Seminars
DESCRIPTION:Prof. Huan Xu\, Georgia Tech\n"The online saddle point problem
  and its applications"\n\nAbstract: \nWe study the online saddle point pro
 blem\, an online learning problem where at each iteration a pair of action
 s need to be chosen without knowledge of the current and future payoff fun
 ctions. The objective is to minimize the gap between the cumulative payoff
 s and the saddle point value of the aggregate payoff function\, which we m
 easure using a metric called “SP-regret”. This problem can be interpre
 ted as finding the Nash equilibrium for the aggregate of a sequence of two
 -player zero-sum game. We propose an algorithm that achieves O(\\sqrt{T}) 
 SP-regret in the general case\, and O(log(T)) SP-regret for the strongly c
 onvex-concave case. Based on these generic results\, we then consider two 
 specific applications\, namely\, online convex optimization with knapsacks
  problem\, and zero-sum matrix game.\n\n\nBio: \nHuan Xu is a senior staff
  engineer at Damo Academy of Alibaba Inc. He received his Ph. D. from McGi
 ll University in 2009. After a postdoc stint at UT-Austin\, he was an assi
 stant professor and then an associate professor (with tenure) at the Natio
 nal University of Singapore. He was an assistant professor at Georgia Inst
 itute of Technology\, before joining Alibaba. He is interested in machine 
 learning and high-dimensional statistics\, robust optimization\, and reinf
 orcement learning. He has published over 90 papers in premier venues in op
 erations research and machine learning. He is currently an associate edito
 r of Mathematics of Operations Research\, IEEE PAMI\, and is an area chair
  of ICML and NIPS.\n\n 
LOCATION:ODY 4 03 https://plan.epfl.ch/?room==ODY%204%2003
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR