Adaptive Moments Are Surprisingly Effective for Plug-and-Play Diffusion Sampling
Abstract
Guided diffusion sampling relies on approximating often intractable likelihood scores, which introduces significant noise into the sampling dynamics. We propose using adaptive moment estimation to stabilize these noisy likelihood scores during sampling. Despite its simplicity, our approach achieves state-of-the-art results on image restoration and class-conditional generation tasks, outperforming more complicated methods, which are often computationally more expensive. We provide empirical analysis of our method on both synthetic and real data, demonstrating that mitigating gradient noise through adaptive moments offers an effective way to improve alignment.
Cite
Text
Belardi et al. "Adaptive Moments Are Surprisingly Effective for Plug-and-Play Diffusion Sampling." International Conference on Learning Representations, 2026.Markdown
[Belardi et al. "Adaptive Moments Are Surprisingly Effective for Plug-and-Play Diffusion Sampling." International Conference on Learning Representations, 2026.](https://mlanthology.org/iclr/2026/belardi2026iclr-adaptive/)BibTeX
@inproceedings{belardi2026iclr-adaptive,
title = {{Adaptive Moments Are Surprisingly Effective for Plug-and-Play Diffusion Sampling}},
author = {Belardi, Christian and Lovelace, Justin and Weinberger, Kilian Q and Gomes, Carla P},
booktitle = {International Conference on Learning Representations},
year = {2026},
url = {https://mlanthology.org/iclr/2026/belardi2026iclr-adaptive/}
}