BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Optimal Reinforcement Learning for Gaussian Systems - Philipp Henn
 ig\, Max Planck Institute for Intelligent Systems\, Department of Empirica
 l Inference\, Tübingen\, Germany
DTSTART:20110919T140000Z
DTEND:20110919T150000Z
UID:TALK33083@talks.cam.ac.uk
CONTACT:Carl Edward Rasmussen
DESCRIPTION:The exploration-exploitation trade-off is among the central ch
 allenges of reinforcement learning. The optimal Bayesian solution is intra
 ctable in general. In this talk I will show that\, however\, if all belief
 s are Gaussian processes\, it *is* possible to make analytic statements ab
 out optimal learning of both rewards and transition dynamics\, for nonline
 ar\, time-varying systems in continuous time and space\, subject to a rela
 tively weak restriction on the dynamics: The solution is described by an i
 nfinite-dimensional partial differential equation. An approximate finite-d
 imensional projection provides a first impression for how this result may 
 be helpful.
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR