BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Statistics
SUMMARY:CCIMI Seminar: Kernel-based Methods for Bandit Con
vex Optimization - Sébastien Bubeck (Microsoft Res
earch Redmond)
DTSTART;TZID=Europe/London:20171004T160000
DTEND;TZID=Europe/London:20171004T170000
UID:TALK84841AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/84841
DESCRIPTION:A lot of progress has been made in recent years on
extending classical multi-armed bandit strategies
to very large set of actions. A particularly chal
lenging setting is the one where the action set is
continuous and the underlying cost function is co
nvex\, this is the so-called bandit convex optimiz
ation (BCO) problem. I will tell the story of BCO
and explain some of the new ideas that we recently
developed to solve it. I will focus on three new
ideas from our recent work http://arxiv.org/abs/16
07.03084 with Yin Tat Lee and Ronen Eldan: (i) a n
ew connection between kernel methods and the popul
ar multiplicative weights strategy\; (ii) a new co
nnection between kernel methods and one of Erdos’
favorite mathematical object\, the Bernoulli convo
lution\, and (iii) a new adaptive (and increasing!
) learning rate for multiplicative weights. These
ideas could be of broader interest in learning/alg
orithm’s design.\n\nThis talk is part of the CCIMI
Seminar Series
LOCATION:MR12
CONTACT:Quentin Berthet
END:VEVENT
END:VCALENDAR