BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Some recent developments in approximate inference: learning and co
 ntrol - David Barber\, University College London
DTSTART:20110706T130000Z
DTEND:20110706T140000Z
UID:TALK31717@talks.cam.ac.uk
CONTACT:Microsoft Research Cambridge Talks Admins
DESCRIPTION:I'll discuss two pieces of work on inference in probabilistic 
 models:\n\nThe first concerns a very general class of Bayesian Linear Mode
 ls that are widely used in statistics and machine learning. A great deal o
 f research has been carried out on developing approximate inference techni
 ques for this important class. In particular I'll discuss methods that bou
 nd the model likelihood\, which is of interest in parameter learning.  The
  well-known `local' variational methods  lower bound the marginal likeliho
 od. Despite their popularity over the last decade\, I'll discuss our recen
 t result that shows that local methods result in weaker bounds than altern
 ative `mean-field' variational methods. In addition\, I'll discuss the per
 haps surprising result that the mean-field bound is concave and discuss ho
 w one may make computationally efficient approximations in large-scale mod
 els with many thousands of variables.\n\nLagrange Duality is being increas
 ingly exploited across machine learning but to date has received comparati
 vely little attention in planning and control. For the second part of the 
 talk I'll discuss an application of Lagrange Duality in learning Markov De
 cision Process policies. In particular\, I'll discuss the computationally 
 difficult finite-horizon time-independent policy case\, and demonstrate ho
 w our method exhibits substantially improved performance compared to polic
 y gradients and more recent `EM' style procedures.
LOCATION:Small lecture theatre\, Microsoft Research Ltd\, 7 J J Thomson Av
 enue (Off Madingley Road)\, Cambridge
END:VEVENT
END:VCALENDAR