Adaptive Moments are Surprisingly Effective for Plug-and-Play Diffusion Sampling

TL;DR

This paper introduces a simple adaptive moment estimation technique to stabilize noisy likelihood scores in guided diffusion sampling, leading to state-of-the-art results in image tasks.

Contribution

The paper presents a novel use of adaptive moments to improve the stability and performance of guided diffusion sampling methods.

Findings

Achieves state-of-the-art results on image restoration and class-conditional generation.

Outperforms more complex, computationally expensive methods.

Effectively mitigates gradient noise in sampling dynamics.

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.