- This event has passed.
LANS Informal Seminar: Brian Van Koten
May 17, 2017 @ 15:00 CDT
Seminar Title: Stratified Markov chain Monte Carlo
Speaker: Brian Van Koten, Department of Statistics, University of Chicago
Date/Time: 2017-05-17 15:00
Location: Building 240, room 4301
Description:
Stratified Markov chain Monte Carlo methods adapt the principle of stratified survey sampling to Markov chain Monte Carlo (MCMC) simulation: In these methods, one divides the sample space into regions called strata, uses MCMC to sample (in parallel) from distributions supported in each of the strata, and finally assembles estimates of general averages from averages over the strata. Stratified MCMC methods arose in chemistry, where they are widely used to compute free energy differences. In this lecture, I will explain how to use stratified MCMC for a much broader class of problems in statistics and computational statistical physics, including both the computation of averages with respect to an arbitrary target distribution and the computation of dynamical quantities such as rates of chemical reactions. I will then present theoretical results and numerical experiments which demonstrate the advantages of stratified MCMC.