BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Variational Gibbs Sampling - Ulrich Paquet
DTSTART:20050427T140000Z
DTEND:20050427T150000Z
UID:TALK4328@talks.cam.ac.uk
CONTACT:Phil Cowans
DESCRIPTION:I introduce a MCMC method for sampling from latent variable \n
 models. The sampling scheme circumvents the traditional latent variable \n
 sample by creating a transition kernel with the required parameter \nposte
 rior as its invariant distribution\, hoping to smooth over local maxima \n
 and trapping states in the latent variable space. In general this kernel i
 s \nnot analytically tractable and I approximate it with a simpler distrib
 ution \nusing an EM bound\; I'll also discuss methods to correct this appr
 oximate \nchain. Finally I'll relate the method to two-stage Gibbs samplin
 g\, EM and \nvariational methods.
LOCATION:Ryle Seminar Room\, Cavendish Laboratory
END:VEVENT
END:VCALENDAR