Adaptive Moments are Surprisingly Effective for Plug-and-Play Diffusion Sampling
Christian Belardi, Justin Lovelace, Kilian Q. Weinberger, Carla P. Gomes · 2026
Read Adaptive Moments are Surprisingly Effective for Plug-and-Play Diffusion Sampling by Christian Belardi, Justin Lovelace, Kilian Q. Weinberger, Carla P. Gomes (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-2603-16797.txt.
Description
Belardi and co-authors stabilize guided diffusion sampling with a simple trick from optimization. Guided sampling must approximate likelihood scores that are hard to compute, and the errors add noise to each step. The team applies adaptive moment estimation, as in the Adam optimizer, to smooth these noisy scores. They report state-of-the-art results on image restoration and class-conditional generation. The method also beats costlier rivals. Tests on synthetic and real data show that cutting gradient noise improves alignment.
Catalog record
| CREATOR | Christian Belardi, Justin Lovelace, Kilian Q. Weinberger, Carla P. Gomes |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 44263 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.16797