Accelerating Convergence of Score-Based Diffusion Models, Provably

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Gen, Huang, Yu, Efimov, Timofey, Wei, Yuting, Chi, Yuejie, Chen, Yuxin
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916149073018880
author Li, Gen
Huang, Yu
Efimov, Timofey
Wei, Yuting
Chi, Yuejie
Chen, Yuxin
author_facet Li, Gen
Huang, Yu
Efimov, Timofey
Wei, Yuting
Chi, Yuejie
Chen, Yuxin
contents Score-based diffusion models, while achieving remarkable empirical performance, often suffer from low sampling speed, due to extensive function evaluations needed during the sampling phase. Despite a flurry of recent activities towards speeding up diffusion generative modeling in practice, theoretical underpinnings for acceleration techniques remain severely limited. In this paper, we design novel training-free algorithms to accelerate popular deterministic (i.e., DDIM) and stochastic (i.e., DDPM) samplers. Our accelerated deterministic sampler converges at a rate $O(1/{T}^2)$ with $T$ the number of steps, improving upon the $O(1/T)$ rate for the DDIM sampler; and our accelerated stochastic sampler converges at a rate $O(1/T)$, outperforming the rate $O(1/\sqrt{T})$ for the DDPM sampler. The design of our algorithms leverages insights from higher-order approximation, and shares similar intuitions as popular high-order ODE solvers like the DPM-Solver-2. Our theory accommodates $\ell_2$-accurate score estimates, and does not require log-concavity or smoothness on the target distribution.
format Preprint
id arxiv_https___arxiv_org_abs_2403_03852
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Accelerating Convergence of Score-Based Diffusion Models, Provably
Li, Gen
Huang, Yu
Efimov, Timofey
Wei, Yuting
Chi, Yuejie
Chen, Yuxin
Machine Learning
Artificial Intelligence
Information Theory
Optimization and Control
Score-based diffusion models, while achieving remarkable empirical performance, often suffer from low sampling speed, due to extensive function evaluations needed during the sampling phase. Despite a flurry of recent activities towards speeding up diffusion generative modeling in practice, theoretical underpinnings for acceleration techniques remain severely limited. In this paper, we design novel training-free algorithms to accelerate popular deterministic (i.e., DDIM) and stochastic (i.e., DDPM) samplers. Our accelerated deterministic sampler converges at a rate $O(1/{T}^2)$ with $T$ the number of steps, improving upon the $O(1/T)$ rate for the DDIM sampler; and our accelerated stochastic sampler converges at a rate $O(1/T)$, outperforming the rate $O(1/\sqrt{T})$ for the DDPM sampler. The design of our algorithms leverages insights from higher-order approximation, and shares similar intuitions as popular high-order ODE solvers like the DPM-Solver-2. Our theory accommodates $\ell_2$-accurate score estimates, and does not require log-concavity or smoothness on the target distribution.
title Accelerating Convergence of Score-Based Diffusion Models, Provably
topic Machine Learning
Artificial Intelligence
Information Theory
Optimization and Control
url https://arxiv.org/abs/2403.03852