BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Variational Inference for Non-Conjugate Models - Dr Guillaume Bouc
 hard (Xerox)
DTSTART:20111026T100000Z
DTEND:20111026T110000Z
UID:TALK33878@talks.cam.ac.uk
CONTACT:Zoubin Ghahramani
DESCRIPTION:Many statistical techniques\, such as the computation of the d
 ata likelihood in the presence of nuisance parameters\, or the prediction 
 in the presence of missing data\, the computation of the posterior distrib
 ution over parameters can be simply expressed as high dimensional integrat
 ion problems for which standard numerical approximation tools cannot be\nd
 irectly applied. The first part of the talk will introduce Split Variation
 al Inference\, a generic way of computing large scale non-Gaussian integra
 ls by splitting them into a sum of small pieces that are easier to approxi
 mate by unnormalized Gaussian distributions. This leads to an any-time imp
 roving algorithm that can be viewed as a generalization of mixture-mean-fi
 eld (variational algorithm where the approximating family is a mixture).\n
 The second part of the talk will present recent developments on the use of
  variational bounds to solve large scale factor analysis/matrix factorizat
 ion problems when data are heterogeneous (i.e. when there are both discret
 e and continuous observations) and heteroscedastic (i.e. when the data var
 iance is not the same for all the observed entities).\n \nCollaborators in
 volved in these works include Onno Zoeter\, Matthias Seeger\, Cedric Archa
 mbeau\, Balaji Lakshminarayanan\, Emtiyaz Khan\, Ben Marlin and Kevin Murp
 hy.\n 
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR