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:Machine Learning Reading Group @ CUED
SUMMARY:GP-BUCB for Spinal Cord Injury Therapy: Batch Acti
ve Learning with Applications - Thomas Desautels
(California Institute of Technology)
DTSTART;TZID=Europe/London:20120620T113000
DTEND;TZID=Europe/London:20120620T120000
UID:TALK38635AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/38635
DESCRIPTION:Can one parallelize complex exploration--exploitat
ion tradeoffs? As an example\, consider the proble
m of optimal high-throughput experimental design\,
where we wish to sequentially design batches of e
xperiments in order to simultaneously learn a surr
ogate function mapping stimulus to response and id
entify the maximum of the function. We formalize
the task as a multi-armed bandit problem\, where t
he unknown payoff function is sampled from a Gauss
ian process (GP)\, and instead of a single arm\,
in each round we pull a batch of several arms in p
arallel. We develop GP-BUCB\, a principled algori
thm for choosing batches\, based on the GP-UCB alg
orithm for sequential GP optimization. We prove a
surprising result\; as compared to the sequential
approach\, the cumulative regret of the parallel
algorithm only increases by a constant factor inde
pendent of the batch size B. Our results provide
rigorous theoretical support for exploiting parall
elism in Bayesian global optimization. We demonstr
ate the effectiveness of our approach on two real-
world applications.
LOCATION:Engineering Department\, CBL Room 438
CONTACT:Konstantina Palla
END:VEVENT
END:VCALENDAR