BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:An Introduction to Variational Methods for Approximate Inference i
 n Graphical Models - Silvia Chiappa
DTSTART:20100217T170000Z
DTEND:20100217T180000Z
UID:TALK22381@talks.cam.ac.uk
CONTACT:Richard Samworth
DESCRIPTION:Many graphical models of practical interest do not admit exact
  probabilistic\ninference and therefore require the use of approximations.
  Variational\nmethods are deterministic approximation methods that have be
 en extensively\nstudied and applied in the Machine Learning community as a
 n alternative to\nstochastic approximation methods like Markov Chain Monte
  Carlo (MCMC)\nmethods. Some of the characteristics that make variational 
 methods more\nattractive than MCMC methods are their often preferable comp
 utational cost\nand their ability to provide bounds on distributions. Whil
 st in theory they\ncannot generate exact results since they are based on a
 n analytical\napproximation to the distribution of interest\,  they have p
 roven to give\nsimilar or superior performance to MCMC methods in several 
 real-world\napplications. In this talk I will explain through simple examp
 les the basic\nprinciples and properties of variational methods and presen
 t some successful\napplications. I will also show how loopy belief propaga
 tion can be\nformulated in a variational framework and introduce a few ext
 ensions that\nhave been derived using this viewpoint. I will finally descr
 ibe the link\nbetween variational transformations and convex duality.\n
LOCATION:MR5\, CMS
END:VEVENT
END:VCALENDAR