Loading [MathJax]/jax/element/mml/optable/GeneralPunctuation.js

2503.21334

Total: 1

#1 Safety of particle filters: Some first results on the time evolution of particle filter estimates [PDF] [Copy] [Kimi] [REL]

Author: Mathieu Gerber

Particle filters (PFs) is a class of Monte Carlo algorithms that propagate over time a set of NN particles which can be used to estimate, in an online fashion, the sequence of filtering distributions (ˆηt)t1 defined by a state-space model. Despite the popularity of PFs, the time evolution of their estimates does not appear to have been previously studied in the literature. Denoting by (ˆηNt)t1 the PF estimate of (ˆηt)t1 and letting κ(0,1), we first show that for any number of particles N it holds that, with probability one, we have for infinitely many t\geq 1, with \|\cdot\| a measure of distance between probability distributions. Considering a simple filtering problem we then provide reassuring results concerning the ability of PFs to estimate jointly a finite set \{\hat{\eta}_t\}_{t=1}^T of filtering distributions by studying \P(\sup_{t\in\{1,\dots,T\}}\|\hat{\eta}_t^{N}-\hat{\eta}_t\|\geq \kappa). Finally, on the same toy filtering problem, we prove that sequential quasi-Monte Carlo, a randomized quasi-Monte Carlo version of PF algorithms, offers greater safety guarantees than PFs in the sense that, for this algorithm, it holds that \lim_{N\rightarrow\infty}\sup_{t\geq 1}\|\hat{\eta}_t^N-\hat{\eta}_t\|=0 with probability one.

Subject: Statistics Theory

Publish: 2025-03-27 10:13:00 UTC