Slowly Annealed Langevin Dynamics: Theory and Applications to Training-Free Guided Generation

TL;DR

This paper introduces Slowly Annealed Langevin Dynamics (SALD), a method for tracking evolving distributions with convergence guarantees, and extends it to Velocity-Aware SALD (VA-SALD) for training-free guided generation using pretrained models.

Contribution

It develops a theoretical framework for SALD with convergence guarantees and introduces VA-SALD for improved guided generation without additional training.

Findings

SALD converges with non-asymptotic guarantees via KL inequalities.

VA-SALD effectively incorporates guidance bias and distribution information.

The framework applies to diffusion-based generative models and similar families.

Abstract

We study Slowly Annealed Langevin Dynamics (SALD), a sampler for tracking a path of moving target distributions and approximating the terminal target through time slowdown. We establish non-asymptotic convergence guarantees via a KL differential inequality, showing that slowdown improves tracking through contraction of intermediate targets and the complexity of the path. Motivated by training-free guided generation with pretrained score-based generative models, we further introduce Velocity-Aware SALD (VA-SALD), which explicitly incorporates the underlying marginal distributions of the pretrained model and uses slowdown to correct the additional deviation induced by guidance. This yields a principled framework for training-free guided generation for diffusion-based and related generative model families, together with convergence guarantees that clarify the roles of intermediate functional inequalities and guidance bias. Code is available at https://github.com/anitan0925/sald.