An Adaptive Particle Allocation Scheme for Offline Sequential Importance Sampling Algorithms

• Activity Number: 661
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Anindya Bhadra*+ and Edward L. Ionides
• Companies: Texas A & M University and University of Michigan
• Keywords: Adaptive particle allocation ; Particle filters ; Sequential Monte Carlo

Abstract

In presence of extreme observations, evaluation of conditional likelihood by a Sequential Monte Carlo filter (also known as a particle filter) suffers from high variance at the corresponding time points, resulting in poor Monte Carlo estimate of the overall likelihood. This paper develops a novel strategy for particle allocation for off-line Sequential Monte Carlo (SMC) based filters, in order to reduce the overall variance of the likelihood estimate. The complications arising from the intractability of the actual SMC variance is handled indirectly by modeling the conditional log likelihood of the observations as an autorgressive process. We demonstrate numerical results on both simulated and real data sets where adaptive particle allocation results in close to 50% lower overall variance over the naive equal allocation of particles at all time points. The technique presented in this article is quite general and applicable to many off-line SMC based optimization techniques.