Uli Schneider
Postdoctoral Researcher, NCAR
 
Monday, 29 September 2003
4:10 PM
E202 Engineering
  
ABSTRACT
Perfect sampling algorithms are Markov Chain Monte Carlo methods without statistical error - if applicable, they enable exact simulation from the stationary distribution on a Markov chain.

We give an introduction to the general idea of perfect sampling and discuss current challenges in this area. We present some advances including "slice coupling" and variants on the so-called IMH-algorithm.

Applications of these advances are given to Bayesian variable selection where one needs to sample from the posterior distribution of the Bayesian model. We employ slice coupling methods and the bounded IMH-algorithm to specify perfect sampling algorithms for selection models in a linear regression setup.

Refreshments will be served at 3:45 p.m. in Room 008 of the Statistics Building