Adaptive Stochastic Coefficients for Accelerating Diffusion Sampling

TL;DR

This paper introduces AdaSDE, a novel adaptive SDE solver that dynamically balances error correction and efficiency, significantly improving diffusion sampling speed and quality across multiple datasets.

Contribution

We propose AdaSDE, a single-step learnable SDE solver that unifies the strengths of ODE and SDE methods through dynamic error regulation, enhancing diffusion sampling performance.

Findings

Achieves state-of-the-art FID scores at 5 NFE on CIFAR-10, FFHQ, and LSUN datasets.

Effectively integrates with existing solvers to improve efficiency and sample quality.

Demonstrates theoretical insights into the weaknesses of ODE and SDE solvers.

Abstract

Diffusion-based generative processes, formulated as differential equation solving, frequently balance computational speed with sample quality. Our theoretical investigation of ODE- and SDE-based solvers reveals complementary weaknesses: ODE solvers accumulate irreducible gradient error along deterministic trajectories, while SDE methods suffer from amplified discretization errors when the step budget is limited. Building upon this insight, we introduce AdaSDE, a novel single-step SDE solver that aims to unify the efficiency of ODEs with the error resilience of SDEs. Specifically, we introduce a single per-step learnable coefficient, estimated via lightweight distillation, which dynamically regulates the error correction strength to accelerate diffusion sampling. Notably, our framework can be integrated with existing solvers to enhance their capabilities. Extensive experiments demonstrate state-of-the-art performance: at 5 NFE, AdaSDE achieves FID scores of 4.18 on CIFAR-10, 8.05 on FFHQ and 6.96 on LSUN Bedroom. Codes are available in https://github.com/WLU-wry02/AdaSDE.