Preconditioned Regularized Wasserstein Proximal Sampling
Abstract. 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.
Authors
- Stanley Osher (ORCID: https://orcid.org/0000-0002-7900-4658)
- Wuchen Li (ORCID: https://orcid.org/0000-0002-2218-5734)
- Hong Ye Tan (ORCID: https://orcid.org/0000-0001-5057-7535)
Institutions
- University of South Carolina (US)
- University of California, Los Angeles (US)
Publication Details
- Journal
- SIAM/ASA Journal on Uncertainty Quantification
- Published
- 2026-11-16
- DOI
- https://doi.org/10.1137/25m1794097
- Citations
- 2
- Primary Topic
- Markov Chains and Monte Carlo Methods
- Type
- article
- Field-Weighted Citation Impact
- 2.92
Funders
- National Science Foundation
- Masason Foundation
- Division of Mathematical Sciences
- Air Force Office of Scientific Research