Sampling from multi-modal distributions on Riemannian manifolds with training-free stochastic interpolants

Alain Durmus, Maxence Noble, Thibaut Pellerin · 2026

Read Sampling from multi-modal distributions on Riemannian manifolds with training-free stochastic interpolants by Alain Durmus, Maxence Noble, Thibaut Pellerin (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-2602-00641.txt.

Description

Alain Durmus and colleagues tackle sampling from unnormalised densities on curved manifolds, where several modes defeat existing methods. Their training-free algorithm moves an easy noise sample toward the target through a deterministic flow out of equilibrium, following an interpolant shaped by the manifold geometry. It rests only on standard Monte Carlo posterior steps, so no learned model is needed, and it carries diffusion-style sampling beyond flat spaces. Theory backs the method, and tests on varied multi-modal cases, including high-dimensional and heavy-tailed targets, show it works.

Catalog record

CREATORAlain Durmus, Maxence Noble, Thibaut Pellerin
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE147158
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.00641

Browse the full library · About the Tower of Babel