BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Assessing high-dimensional latent variable models - Dr Iain Murray
  (Toronto)
DTSTART:20080612T130000Z
DTEND:20080612T140000Z
UID:TALK12399@talks.cam.ac.uk
CONTACT:Zoubin Ghahramani
DESCRIPTION:Having built a probabilistic model\, a natural question is:\n
 “what probability does my model assign to the data?”.\n\nWe might fit 
 the model’s parameters to avoid having to compute an intractable margina
 l likelihood. Even then\, evaluating a test-set probability with fixed par
 ameters can be difficult. I will discuss recent work on evaluating high-di
 mensional undirected graphical models and models with many latent variable
 s. This allows direct comparisons of the probabilistic predictions made by
  graphical models with hundreds of thousands of parameters against simpler
  alternatives.\n\nDetails: I will review Restricted Boltzmann Machines (RB
 M’s) and Annealed Importance Sampling (AIS). Then I will present new wor
 k that allows efficient assessment of more general Boltzmann machines and 
 Deep Belief Networks.
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR