BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Learning Conditional Random Fields with Hierarchical Features: App
 lication to the Game of Go - Scott Sanner\, University of Toronto
DTSTART:20061129T150000Z
DTEND:20061129T160000Z
UID:TALK5995@talks.cam.ac.uk
CONTACT:Oliver Williams
DESCRIPTION:We examine an important subtask of policy learning in the game
  of Go: approximating the value function given a fixed policy. We model th
 e value function as the expected territory outcome of a Go board configura
 tion and learn to predict this outcome using a conditional Markov random f
 ield (CRF). This task is complicated by the complexity of inference on a G
 o Board (361 individual territories to predict – all influenced by surro
 unding positions) and the use of 4 million pattern-based features. Such co
 mplexity induces many computational and statistical problems which must be
  accounted for during both training and inference. \nIn this work we exami
 ne a variety of models (Independent vs. Coupled\, Flat vs. Hierarchical)\,
  learning algorithms (Local Training vs. Max Likelihood vs. Max Pseudo-lik
 elihood)\, and inference approaches (Loopy BP vs. Sampling\, Bayesian Mode
 l Averaging vs. Heuristic Model Selection). We present results from learni
 ng to predict territory in expert games and conclude with a prescription f
 or future work on approximating the value function in Go. \nThis is joint 
 work with Thore Graepel and Ralf Herbrich with contributions by Tom Minka.
 \n
LOCATION:Small public lecture room\, Microsoft Research Ltd\, 7 J J Thomso
 n Avenue (Off Madingley Road)\, Cambridge
END:VEVENT
END:VCALENDAR