Auxiliary MCMC samplers for parallelisable inference in high-dimensional latent dynamical systems

Loading...
Thumbnail Image

Access rights

openAccess
CC BY
publishedVersion

URL

Journal Title

Journal ISSN

Volume Title

A1 Alkuperäisartikkeli tieteellisessä aikakauslehdessä

Date

Major/Subject

Mcode

Degree programme

Language

en

Pages

55

Series

Electronic Journal of Statistics, Volume 19, issue 1, pp. 1370-1424

Abstract

Sampling from the full posterior distribution of high-dimen-sional non-linear, non-Gaussian latent dynamical models presents significant computational challenges. While Particle Gibbs (also known as conditional sequential Monte Carlo) is considered the gold standard for this task, it quickly degrades in performance as the latent space dimensionality increases. Conversely, globally Gaussian-approximated methods like extended Kalman filtering, though more robust, are seldom used for posterior sampling due to their inherent bias. We introduce novel auxiliary sampling approaches that address these limitations. By incorporating arti-ficial observations of the system as auxiliary variables in our MCMC ker-nels, we develop both efficient exact Kalman-based samplers and enhanced Particle Gibbs algorithms that maintain performance in high-dimensional latent spaces. Some of our methods support parallelization along the time dimension, achieving logarithmic scaling when implemented on GPUs. Empirical evaluations demonstrate superior statistical and computational performance compared to existing approaches for high-dimensional latent dynamical systems.

Description

Publisher Copyright: © 2025, Institute of Mathematical Statistics. All rights reserved.

Other note

Citation

Corenflos, A & Särkkä, S 2025, 'Auxiliary MCMC samplers for parallelisable inference in high-dimensional latent dynamical systems', Electronic Journal of Statistics, vol. 19, no. 1, pp. 1370-1424. https://doi.org/10.1214/25-EJS2363