Multilevel Stein variational gradient descent with applications to Bayesian inverse problems

Terrence Alsup, Luca Venturi, Benjamin Peherstorfer

Research output: Contribution to journalConference articlepeer-review

Abstract

This work presents a multilevel variant of Stein variational gradient descent to more efficiently sample from target distributions. The key ingredient is a sequence of distributions with growing fidelity and costs that converges to the target distribution of interest. For example, such a sequence of distributions is given by a hierarchy of ever finer discretization levels of the forward model in Bayesian inverse problems. The proposed multilevel Stein variational gradient descent moves most of the iterations to lower, cheaper levels with the aim of requiring only a few iterations on the higher, more expensive levels when compared to the traditional, single-level Stein variational gradient descent variant that uses the highest-level distribution only. Under certain assumptions, in the mean-field limit, the error of the proposed multilevel Stein method decays by a log factor faster than the error of the single-level counterpart with respect to computational costs. Numerical experiments with Bayesian inverse problems show speedups of more than one order of magnitude of the proposed multilevel Stein method compared to the single-level variant that uses the highest level only.

Original languageEnglish (US)
Pages (from-to)93-117
Number of pages25
JournalProceedings of Machine Learning Research
Volume145
StatePublished - 2021
Event2nd Mathematical and Scientific Machine Learning Conference, MSML 2021 - Virtual, Online
Duration: Aug 16 2021Aug 19 2021

Keywords

  • Bayesian inference
  • Monte Carlo
  • multilevel and multifidelity
  • particle methods

ASJC Scopus subject areas

  • Artificial Intelligence
  • Software
  • Control and Systems Engineering
  • Statistics and Probability

Fingerprint

Dive into the research topics of 'Multilevel Stein variational gradient descent with applications to Bayesian inverse problems'. Together they form a unique fingerprint.

Cite this