BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Bayesian Reinforcement Learning - Rowan McAllister &amp\; Karolina
  Dziugaite
DTSTART:20130321T150000Z
DTEND:20130321T163000Z
UID:TALK43633@talks.cam.ac.uk
CONTACT:Colorado Reed
DESCRIPTION:Reinforcement learning (RL) is the problem of learning optimal
  behaviour in an initially unfamiliar Markov Decision Process (MDP) enviro
 nment through interaction and evaluative feedback.\nUntil recently\, exist
 ing RL algorithms have relied on non-optimal exploration strategies to str
 ike a balance between 'exploiting' current knowledge of the MDP to maximis
 e expected returns\, and 'exploration' actions which gain information on t
 he MDP\, to improve the return on exploitation actions in the future.\nBay
 esian Reinforcement learning (BRL) is about capturing and dealing with unc
 ertainty in MDP elements\, where 'classic RL' does not.\nWe focus on model
 ling uncertainty in an agent's transition probabilities\, often termed 'mo
 del-based' BRL.\nBy planning in a belief space of transition probabilities
 \, BRL implicitly resolves the classic RL 'exploitation & exploration' dil
 emma optimally.\nComputation is shown to be intractable in general\, altho
 ugh approximations exist of which several key algorithms are presented.
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR