BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Adaptive Monte Carlo Methods for Simulation and Optimization - Chu
 nlin JI\, Engineering Department
DTSTART:20050817T140000Z
DTEND:20050817T150000Z
UID:TALK4426@talks.cam.ac.uk
CONTACT:Phil Cowans
DESCRIPTION:When using Monte Carlo methods\, it is of interest to adapt th
 e proposal density  to obtain faster convergence and more accurate estimat
 ion. In order to adapt the proposal\, a Cross Entropy criterion is propose
 d\, with some adaptive strategies based on Stochastic Approximation. A nov
 el adaptive population MCMC algorithm is addressed. And how to use a mixtu
 re of distribution as proposal is also discussed. Moreover\, these adaptat
 ions are also suitable for Importance Sampling. \n\nThese adaptive Monte C
 arlo methods are able to be optimization approaches by using some annealin
 g scheme. To show the advantages\, they are compared with related methods\
 , e.g. Simulated Annealing. \n\nFor Bayesian Inference\, a reversible jump
  version of adaptive Monte Carlo method is proposed to perform parameter e
 stimation/optimization and model selection simultaneously. Variational met
 hod with above adaptive strategies forms another approach for approximatin
 g inference and learning under Bayesian model.
LOCATION:Ryle Seminar Room\, Cavendish Laboratory
END:VEVENT
END:VCALENDAR