Particle algorithms for maximum likelihood training of latent variable models
Computation
2023-02-21 v5 Machine Learning
Optimization and Control
Methodology
Machine Learning
Abstract
(Neal and Hinton, 1998) recast maximum likelihood estimation of any given latent variable model as the minimization of a free energy functional , and the EM algorithm as coordinate descent applied to . Here, we explore alternative ways to optimize the functional. In particular, we identify various gradient flows associated with and show that their limits coincide with 's stationary points. By discretizing the flows, we obtain practical particle-based algorithms for maximum likelihood estimation in broad classes of latent variable models. The novel algorithms scale to high-dimensional settings and perform well in numerical experiments.
Cite
@article{arxiv.2204.12965,
title = {Particle algorithms for maximum likelihood training of latent variable models},
author = {Juan Kuntz and Jen Ning Lim and Adam M. Johansen},
journal= {arXiv preprint arXiv:2204.12965},
year = {2023}
}