Preconditioned Regularized Wasserstein Proximal Sampling
SIAM/ASA Journal on Uncertainty Quantification · 31.12.2026 · 14 (4) · 1592-1621
DOI: 10.1137/25m1794097 ↗Kurz gefasst
Fachartikel in der Zeitschrift SIAM/ASA Journal on Uncertainty Quantification, online erschienen am 16.11.2026. Verfasst von Hong Ye Tan und 2 weiteren Autor:innen. Die Arbeit nennt 60 Literaturangaben. Gefördert von: Air Force Office of Scientific Research, Division of Mathematical Sciences, National Science Foundation.
Abstract (Originalsprache)
We consider sampling from a Gibbs distribution by evolving finitely many particles. We propose a preconditioned version of a recently proposed noise-free sampling method, governed by approximating the score function with the numerically tractable score of a regularized Wasserstein proximal operator. This is derived by a Cole–Hopf transformation on coupled anisotropic heat equations, yielding a kernel formulation for the preconditioned regularized Wasserstein proximal. The diffusion component of the proposed method is also interpreted as a modified self-attention block, as in transformer architectures. For quadratic potentials, we provide a discrete-time nonasymptotic convergence analysis and explicitly characterize the bias, which is dependent on regularization and independent of step-size. Experiments demonstrate acceleration and particle-level stability on various log-concave and non-log-concave toy examples, as well as for Bayesian total-variation regularized image deconvolution. We further show competitive/better performance on nonconvex Bayesian neural network training when utilizing variable preconditioning matrices.
Förderung
- Air Force Office of Scientific Research
- Division of Mathematical Sciences
- National Science Foundation
Zitieren
APA
Tan, H. Y., Osher, S., & Li, W. (2026). Preconditioned Regularized Wasserstein Proximal Sampling. SIAM/ASA Journal on Uncertainty Quantification, 14(4), 1592-1621. https://doi.org/10.1137/25m1794097
Vancouver
Tan HY, Osher S, Li W. Preconditioned Regularized Wasserstein Proximal Sampling. SIAM/ASA Journal on Uncertainty Quantification. 2026;14(4):1592-1621. doi:10.1137/25m1794097
BibTeX
@article{Tan2026,
author = {Tan, Hong Ye and Osher, Stanley and Li, Wuchen},
title = {Preconditioned Regularized Wasserstein Proximal Sampling},
journal = {SIAM/ASA Journal on Uncertainty Quantification},
year = {2026},
volume = {14},
number = {4},
pages = {1592--1621},
doi = {10.1137/25m1794097}
}
EUVIMED