BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:CCIMI Seminar: Kernel-based Methods for Bandit Convex Optimization
  - Sébastien Bubeck (Microsoft Research Redmond)
DTSTART:20171004T150000Z
DTEND:20171004T160000Z
UID:TALK84841@talks.cam.ac.uk
CONTACT:Quentin Berthet
DESCRIPTION:A lot of progress has been made in recent years on extending c
 lassical multi-armed bandit strategies to very large set of actions. A par
 ticularly challenging setting is the one where the action set is continuou
 s and the underlying cost function is convex\, this is the so-called bandi
 t convex optimization (BCO) problem. I will tell the story of BCO and expl
 ain some of the new ideas that we recently developed to solve it. I will f
 ocus on three new ideas from our recent work http://arxiv.org/abs/1607.030
 84 with Yin Tat Lee and Ronen Eldan: (i) a new connection between kernel m
 ethods and the popular multiplicative weights strategy\; (ii) a new connec
 tion between kernel methods and one of Erdos’ favorite mathematical obje
 ct\, the Bernoulli convolution\, and (iii) a new adaptive (and increasing!
 ) learning rate for multiplicative weights. These ideas could be of broade
 r interest in learning/algorithm’s design.\n\nThis talk is part of the C
 CIMI Seminar Series
LOCATION:MR12
END:VEVENT
END:VCALENDAR