BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:GP-BUCB for Spinal Cord Injury Therapy: Batch Active Learning with
  Applications - Thomas Desautels  (California Institute of Technology)
DTSTART:20120620T103000Z
DTEND:20120620T110000Z
UID:TALK38635@talks.cam.ac.uk
CONTACT:Konstantina Palla
DESCRIPTION:Can one parallelize complex exploration--exploitation tradeoff
 s? As an example\, consider the problem of optimal high-throughput experim
 ental design\, where we wish to sequentially design batches of experiments
  in order to simultaneously learn a surrogate function mapping stimulus to
  response and identify the maximum of the function.  We formalize the task
  as a multi-armed bandit problem\, where the unknown payoff function is sa
 mpled from a Gaussian process (GP)\, and instead of a single arm\,  in eac
 h round we pull a batch of several arms in parallel.  We develop GP-BUCB\,
  a principled algorithm for choosing batches\, based on the GP-UCB algorit
 hm for sequential GP optimization.  We prove a surprising result\; as comp
 ared to the sequential approach\, the cumulative regret of the parallel al
 gorithm only increases by a constant factor independent of the batch size 
 B.  Our results provide rigorous theoretical support for exploiting parall
 elism in Bayesian global optimization. We demonstrate the effectiveness of
  our approach on two real-world applications.
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR