SAMC Particle Filter for Nonlinear State Space Models

• Activity Number: 523
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Mingqi Wu*+ and Faming Liang
• Companies: Texas A&M University and Texas A&M University
• Keywords: Sequential Monte Carlo ; Stochastic approximation Monte Carlo ; Particle filter ; Importance sampling ; Nonlinear state space model

Abstract

Particle filters provide an attractive approach to approximate a sequence of probability distributions of interest using a large population of important samples/particles. One open problem in particle filter research is that the weight of the particles tend to degenerate as the number of filtration steps increases. In this talk, we propose a stochastic approximation Monte Carlo (SAMC) based particle filter, which avoids the notorious weight degeneracy problem by taking advantage of two attractive features of the SAMC algorithm: (i)superiority in sample space exploration and (ii)ability to generate weight-bounded importance samples. We compare the new particle filter with some existing particle filters on two non-linear state space models. The numerical results indicate that the new particle filter can significantly outperform the others.