Adaptive Momentum Langevin Dynamics for Efficient Posterior Sampling
Entropy · 01.08.2026 · 28 (8) · 854
DOI: 10.3390/e28080854 ↗Kurz gefasst
Fachartikel in der Zeitschrift Entropy, online erschienen am 01.08.2026. Verfasst von Zhengbo Li und 3 weiteren Autor:innen. Beteiligte Einrichtung: School of Mathematics and Statistics, Guizhou University, Guiyang 550025, China. Die Arbeit nennt 15 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Key Project of Guizhou Provincial Basic Research Program, Youth Guidance Project of Guizhou Provincial Basic Research Program.
Abstract (Originalsprache)
Sampling from high-dimensional posterior distributions is a central challenge in Bayesian inference and noisy inverse problems. Standard first-order Langevin-based methods often suffer from slow convergence and sensitivity to step-size hyperparameters, particularly in annealed score-based inverse imaging pipelines. We propose Adaptive Momentum Langevin Dynamics (AMLD), a practical stochastic correction kernel that introduces a momentum variable into the annealed posterior sampling framework and equips it with an annealing-aware momentum retention schedule. The method is fully compatible with the SNIPS framework and retains its coordinate-wise adaptive step structure, acting as a lightweight drop-in replacement for the conventional first-order Langevin correction step. Extensive experiments on three representative image inverse problems—Gaussian deblurring, inpainting, and 4× super-resolution—demonstrate that AMLD consistently achieves strong PSNR and LPIPS performance, with competitive FID in most settings, compared to three state-of-the-art baselines (DDRM, DPS, SNIPS) under both nearly noiseless and noisy measurement conditions, while reaching target reconstruction quality using fewer sampling iterations. The proposed momentum-based sampler provides empirically improved exploration and robustness across evolving posterior landscapes, offering a practical and computationally efficient alternative to first-order annealed Langevin samplers in high-dimensional Bayesian inverse problems.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- School of Mathematics and Statistics, Guizhou University, Guiyang 550025, China.
- School of General Education, Moutai Institute, Renhuai 564507, China.
- School of Mathematics, South China University of Technology, Guangzhou 510640, China.
- Guizhou Provincial Laboratory of Big Data, Guizhou University, Guiyang 550025, China.
ORCID-Profile: Zhengbo Li, Jian Xu, Dingtao Peng
Förderung
- National Natural Science Foundation of China
- Key Project of Guizhou Provincial Basic Research Program
- Youth Guidance Project of Guizhou Provincial Basic Research Program
- Research of Zunyi Municipal Science and Technology Plan Project
- Research of Moutai Institute High-Level Talent Research Startup Funding Project
Zitieren
APA
Li, Z., Xu, J., Peng, D., & Hu, S. (2026). Adaptive Momentum Langevin Dynamics for Efficient Posterior Sampling. Entropy, 28(8), 854. https://doi.org/10.3390/e28080854
Vancouver
Li Z, Xu J, Peng D, Hu S. Adaptive Momentum Langevin Dynamics for Efficient Posterior Sampling. Entropy. 2026;28(8):854. doi:10.3390/e28080854
BibTeX
@article{Li2026,
author = {Li, Zhengbo and Xu, Jian and Peng, Dingtao and Hu, Shuang},
title = {Adaptive Momentum Langevin Dynamics for Efficient Posterior Sampling},
journal = {Entropy},
year = {2026},
volume = {28},
number = {8},
pages = {854},
doi = {10.3390/e28080854}
}
EUVIMED