English

A Non-Asymptotic Analysis for Stein Variational Gradient Descent

Machine Learning 2021-01-05 v4 Machine Learning

Abstract

We study the Stein Variational Gradient Descent (SVGD) algorithm, which optimises a set of particles to approximate a target probability distribution πeV\pi\propto e^{-V} on Rd\mathbb{R}^d. In the population limit, SVGD performs gradient descent in the space of probability distributions on the KL divergence with respect to π\pi, where the gradient is smoothed through a kernel integral operator. In this paper, we provide a novel finite time analysis for the SVGD algorithm. We provide a descent lemma establishing that the algorithm decreases the objective at each iteration, and rates of convergence for the average Stein Fisher divergence (also referred to as Kernel Stein Discrepancy). We also provide a convergence result of the finite particle system corresponding to the practical implementation of SVGD to its population version.

Keywords

Cite

@article{arxiv.2006.09797,
  title  = {A Non-Asymptotic Analysis for Stein Variational Gradient Descent},
  author = {Anna Korba and Adil Salim and Michael Arbel and Giulia Luise and Arthur Gretton},
  journal= {arXiv preprint arXiv:2006.09797},
  year   = {2021}
}

Comments

Accepted to Neurips 2020

R2 v1 2026-06-23T16:24:05.432Z