BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Annealing Between Distributions by Averaging Moments - Chris Maddi
 son (U Toronto)
DTSTART:20130802T100000Z
DTEND:20130802T110000Z
UID:TALK46498@talks.cam.ac.uk
CONTACT:Zoubin Ghahramani
DESCRIPTION:Many powerful Monte Carlo techniques for estimating partition 
 functions\, such as annealed importance sampling (AIS)\, are based on samp
 ling from a sequence of intermediate distributions\, which interpolate bet
 ween a tractable initial distribution and an intractable target distributi
 on. The near-universal practice is to use geometric averages of the initia
 l and target distributions\, but alternative paths can perform substantial
 ly better. We present a novel sequence of intermediate distributions for e
 xponential families: averaging the moments of the initial and target distr
 ibutions. We derive an asymptotically optimal piecewise linear schedule fo
 r the moments path and show that it performs at least as well as geometric
  averages with a  linear schedule. Moment averaging performs well empirica
 lly at estimating partition functions of restricted Boltzmann machines (RB
 Ms)\, which form the building blocks of many deep learning models.
LOCATION:Engineering Department\, CBL Room BE-438
END:VEVENT
END:VCALENDAR