Semi-independent resampling for particle filtering

TL;DR

This paper introduces a new class of parameterized SIR-based methods for particle filtering that better balance computational cost and statistical accuracy, especially in high-dimensional or informative models.

Contribution

It proposes a novel parameterized rejuvenation mechanism for SIR algorithms, improving their effectiveness in challenging high-dimensional settings.

Findings

Enhanced performance in high-dimensional models

Flexible adjustment of computational and statistical tradeoffs

Improved particle filter robustness

Abstract

Among Sequential Monte Carlo (SMC) methods,Sampling Importance Resampling (SIR) algorithms are based on Importance Sampling (IS) and on some resampling-based)rejuvenation algorithm which aims at fighting against weight degeneracy. However %whichever the resampling technique used this mechanism tends to be insufficient when applied to informative or high-dimensional models. In this paper we revisit the rejuvenation mechanism and propose a class of parameterized SIR-based solutions which enable to adjust the tradeoff between computational cost and statistical performances.