Bridge Matching Sampler: Scalable Sampling via Generalized Fixed-Point Diffusion Matching
Denis Blessing, Lorenz Richter, Julius Berner, Egor Malitskiy, Gerhard Neumann · 2026
Read Bridge Matching Sampler: Scalable Sampling via Generalized Fixed-Point Diffusion Matching by Denis Blessing, Lorenz Richter, Julius Berner, Egor Malitskiy, Gerhard Neumann (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-2603-00530.txt.
Description
Blessing and four colleagues improve sampling from unnormalised densities with diffusion models, where least squares matching methods force trade offs such as restricted priors or unstable optimisation. Viewing those methods as fixed point iterations rooted in Nelson's relation, they build the Bridge Matching Sampler. It learns a stochastic transport map between any prior and target with one scalable stable objective. A damped variant adds regularisation to fight mode collapse and steady training further. The method samples at unprecedented scale while preserving mode diversity, leading on synthetic and molecular benchmarks.
Catalog record
| CREATOR | Denis Blessing, Lorenz Richter, Julius Berner, Egor Malitskiy, Gerhard Neumann |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 136640 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.00530