Adaptive Momentum Langevin Dynamics for Efficient Posterior Sampling
Zhengbo Li, Jian Xu, Dingtao Peng, Shuang Hu · 2026
Read Adaptive Momentum Langevin Dynamics for Efficient Posterior Sampling by Zhengbo Li, Jian Xu, Dingtao Peng, Shuang Hu (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/PMC13512748.txt.
Description
Li and colleagues proposed Adaptive Momentum Langevin Dynamics, a sampling algorithm for restoring images from partial measurements within a Bayesian framework, in a 2026 methods study. Standard Langevin samplers, which wander through possible solutions guided by chance, converge slowly and react badly to step-size choices inside annealed pipelines that lower noise. The new method adds a momentum term, borrowed from fast search methods.
Li tuned how long momentum persists to the cooling schedule, so search stays stable early and settles late. They designed the update as a drop-in swap inside a score-based framework. The tests spanned three benchmark tasks: unblurring, filling missing regions, and enlarging images fourfold. On these, the sampler matched or beat three state-of-the-art rivals on sharpness and perceived quality scores under both nearly clean and noisy conditions. Removal tests showed the momentum itself, kept strong, drove most of the gain.
Catalog record
| CREATOR | Zhengbo Li, Jian Xu, Dingtao Peng, Shuang Hu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36 KB |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |