On the Convergence of Wasserstein Gradient Descent for Sampling

Van Chien Ta · 2026

Read On the Convergence of Wasserstein Gradient Descent for Sampling by Van Chien Ta (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-03413.txt.

Description

Drawing samples from a complicated probability distribution is a core task in Bayesian statistics. Van Chien Ta and 2 co-authors study a sampling method built on Wasserstein gradient descent in their 2026 paper. The method follows the downhill flow of an energy function, the KL functional, through the space of probability measures itself.

The authors identify two subclasses of that space where the descent scheme is guaranteed to converge. For real use, they build a particle version that estimates the needed score function by score matching. Numerical experiments show good approximations of difficult target distributions, including cases where Markov chain Monte Carlo and variational Bayes struggle. The results support optimization-based sampling for high-dimensional problems.

Catalog record

CREATORVan Chien Ta
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-03413

Browse the full library · About the Tower of Babel