Constrained Sequential Monte Carlo

Abstract

The sequential Monte Carlo (SMC) methodology has shown a great promise in solving a large class of highly complex inference and optimization problems. Originally it is designed to solve on-line filtering and smoothing problems for non-linear non-Gaussian state space models. Recently it has been shown to be equally powerful in dealing with fixed-dimensional problems, utilizing a sequential decomposition principle. In this talk we discuss issues and efficient implementations of SMC for dealing with a class of high dimensional distributions that are defined on restricted and ill-shaped spaces. Examples in bioinformatics and financial engineering are presented. ?