BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Divergence measures and message passing - David Knowles (Universit
 y of Cambridge)
DTSTART:20090514T130000Z
DTEND:20090514T143000Z
UID:TALK18076@talks.cam.ac.uk
CONTACT:Shakir Mohamed
DESCRIPTION:I will present Tom Minka's paper\, "Divergence measures and me
 ssage passing"\n\nThis paper presents a unifying view of message-passing a
 lgorithms\, as methods to approximate a complex Bayesian network by a simp
 ler network with minimum information divergence. In this view\, the differ
 ence between mean-field methods and belief propagation is not the amount o
 f structure they model\, but only the measure of loss they minimize (`excl
 usive' versus `inclusive' Kullback-Leibler divergence). In each case\, mes
 sage-passing arises by minimizing a localized version of the divergence\, 
 local to each factor. By examining these divergence measures\, we can intu
 it the types of solution they prefer (symmetry-breaking\, for example) and
  their suitability for different tasks. Furthermore\, by considering a wid
 er variety of divergence measures (such as alpha-divergences)\, we can ach
 ieve different complexity and performance goals.\n\nhttp://research.micros
 oft.com/en-us/um/people/minka/papers/message-passing/\n
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR