Twisted particle filters

TL;DR

This paper introduces 'twisted' particle filters, a new class of algorithms that optimize sampling laws for particle approximations in hidden Markov models, significantly improving fluctuation properties and efficiency.

Contribution

It characterizes the unique optimal transition kernels for particle systems and proposes a novel twisted particle filter algorithm based on these kernels.

Findings

Optimal transition kernels minimize asymptotic variance growth.

Twisted particle filters outperform standard algorithms in fluctuation properties.

Asymptotic analysis confirms improved efficiency with many particles.

Abstract

We investigate sampling laws for particle algorithms and the influence of these laws on the efficiency of particle approximations of marginal likelihoods in hidden Markov models. Among a broad class of candidates we characterize the essentially unique family of particle system transition kernels which is optimal with respect to an asymptotic-in-time variance growth rate criterion. The sampling structure of the algorithm defined by these optimal transitions turns out to be only subtly different from standard algorithms and yet the fluctuation properties of the estimates it provides can be dramatically different. The structure of the optimal transition suggests a new class of algorithms, which we term "twisted" particle filters and which we validate with asymptotic analysis of a more traditional nature, in the regime where the number of particles tends to infinity.