Adaptive Kernel Density Estimation proposal in gravitational wave data analysis

TL;DR

This paper proposes an adaptive Kernel Density Estimation-based proposal method for Markov Chain Monte Carlo sampling in gravitational wave data analysis, aiming to improve sampling efficiency especially with increasing data volume.

Contribution

It introduces a KDE-based adaptive proposal that groups parameters by correlation and stabilizes over iterations, enhancing MCMC efficiency in gravitational wave data analysis.

Findings

KDE proposal can reduce autocorrelation length in chains.

Method is effective on IPTA and LISA datasets.

Performance decreases with strong parameter correlations.

Abstract

Markov Chain Monte Carlo approach is frequently used within Bayesian framework to sample the target posterior distribution. Its efficiency strongly depends on the proposal used to build the chain. The best jump proposal is the one that closely resembles the unknown target distribution, therefore we suggest an adaptive proposal based on Kernel Density Estimation (KDE). We group parameters of the model according to their correlation and build KDE based on the already accepted points for each group. We adapt the KDE-based proposal until it stabilizes. We argue that such a proposal could be helpful in applications where the data volume is increasing and in the hyper-model sampling. We tested it on several astrophysical datasets (IPTA and LISA) and have shown that in some cases KDE-based proposal also helps to reduce the autocorrelation length of the chains. The efficiency of this proposal is reduces in case of the strong correlations between a large group of parameters.