Fall Seminar Series - September 15, 2005
University of Minnesota
School of Statistics
College of Liberal Arts

Stochastic Approximation in Monte Carlo Computation

Faming Liang
Department of Statistics
Texas A & M University

Thursday, September 15, 2005
3:30 PM, 115 Ford Hall
Minneapolis, East Bank Campus
Social at 3:00 PM, 300 Ford Hall

Abstract

The Wang-Landau algorithm (Wang and Landau, 2001) has received much attention in the literature recently, although the lack of a rigorous theory on its convergence has been a concern. In this talk, I will present the stochastic approximation Monte Carlo (SAMC) algorithm, which can be regarded as a stochastic approximation extension of the Wang-Landau algorithm. The convergence of the algorithm will also be addressed. SAMC represents a new development of the stochastic approximation method (Robbins and Monro, 1951), extending the applications of stochastic approximation to Monte Carlo computation. The numerical results on importance sampling, model selection, and Monte Carlo optimization indicate that with an appropriate setting, SAMC outperforms the methods conventionally used for these problems. This talk is partially based on a joint work with Chuanhai Liu and Raymond J. Carroll.