Adaptive Non-uniform Timestep Sampling for Accelerating Diffusion Model Training

TL;DR

This paper introduces an adaptive non-uniform timestep sampling method for diffusion models that accelerates training and enhances performance by focusing on high-variance, critical timesteps during the learning process.

Contribution

The paper presents a novel adaptive sampling technique that dynamically prioritizes important timesteps, improving training efficiency and robustness across multiple datasets and architectures.

Findings

Accelerates diffusion model training by focusing on critical timesteps.

Improves model performance at convergence.

Demonstrates robustness across datasets and architectures.

Abstract

As a highly expressive generative model, diffusion models have demonstrated exceptional success across various domains, including image generation, natural language processing, and combinatorial optimization. However, as data distributions grow more complex, training these models to convergence becomes increasingly computationally intensive. While diffusion models are typically trained using uniform timestep sampling, our research shows that the variance in stochastic gradients varies significantly across timesteps, with high-variance timesteps becoming bottlenecks that hinder faster convergence. To address this issue, we introduce a non-uniform timestep sampling method that prioritizes these more critical timesteps. Our method tracks the impact of gradient updates on the objective for each timestep, adaptively selecting those most likely to minimize the objective effectively. Experimental results demonstrate that this approach not only accelerates the training process, but also leads to improved performance at convergence. Furthermore, our method shows robust performance across various datasets, scheduling strategies, and diffusion architectures, outperforming previously proposed timestep sampling and weighting heuristics that lack this degree of robustness.