Bayesian Analysis via Stochastic Approximation Monte Carlo for Statistical Model with Intractable Normalizing Constants

• Activity Number: 39
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Ick Hoon Jin* and Faming Liang+
• Companies: Texas A&M University and Texas A&M University
• Keywords: Intractable Normalizing Constant ; Stochastic Approximation Monte Carlo ; Bayesian Parametric Estimation ; Ising Model ; Autologistic Model ; Autonormal Model

Abstract

From now on, Bayesian parametric estimations for models with intractable normalizing constant have been attracted in many literatures. In this paper, we propose a new algorithm, Bayesian Estimation via Stochastic Approximation Monte Carlo (BSAMC), to address intractable normalizing constant problems. At each iteration, we update a normalizing constant by the rule of stochastic approximation Monte Carlo method with pre-defined subregions of energy function. Although our approach needs an initial guess of parameters to construct pre-defined subregions, this algorithm produces consistent results, as illustrated in Ising model examples. For two other models, spatial autologistic model and autonormal model, our algorithm also induces equivalent results with other pre-existing algorithms. As discussed, BSAMC algorithm can be applied to many models with intractable normalizing constants.