How Much is Enough? A Study on Diffusion Times in Score-based Generative Models

Abstract

Score-based diffusion models are a class of generative models whose dynamics is described by stochastic differential equations that map noise into data. While recent works have started to lay down a theoretical foundation for these models, an analytical understanding of the role of the diffusion time T is still lacking. Current best practice advocates for a large T to ensure that the forward dynamics brings the diffusion sufficiently close to a known and simple noise distribution; however, a smaller value of T should be preferred for a better approximation of the score-matching objective and higher computational efficiency. Starting from a variational interpretation of diffusion models, in this work we quantify this trade-off, and suggest a new method to improve quality and efficiency of both training and sampling, by adopting smaller diffusion times. Indeed, we show how an auxiliary model can be used to bridge the gap between the ideal and the simulated forward dynamics, followed by a standard reverse diffusion process. Empirical results support our analysis; for image data, our method is competitive w.r.t. the state-of-the-art, according to standard sample quality metrics and log-likelihood.

References in corpus (16)

• Score-Based Generative Modeling through Stochastic Differential Equations
• Improved Denoising Diffusion Probabilistic Models
• Progressive Distillation for Fast Sampling of Diffusion Models
• Tackling the Generative Learning Trilemma with Denoising Diffusion GANs
• Structured Denoising Diffusion Models in Discrete State-Spaces
• CSDI: Conditional Score-based Diffusion Models for Probabilistic Time Series Imputation
• Analytic-DPM: an Analytic Estimate of the Optimal Reverse Variance in Diffusion Probabilistic Models
• Diffusion Schrödinger Bridge with Applications to Score-Based Generative Modeling
• Learning to Efficiently Sample from Diffusion Probabilistic Models
• Maximum Likelihood Training of Score-Based Diffusion Models
• The Role of ImageNet Classes in Fréchet Inception Distance
• A Variational Perspective on Diffusion-Based Generative Models and Score Matching
• PriorGrad: Improving Conditional Denoising Diffusion Models with Data-Dependent Adaptive Prior
• Gotta Go Fast When Generating Data with Score-Based Models
• Score-Based Generative Modeling with Critically-Damped Langevin Diffusion
• Likelihood Training of Schrödinger Bridge using Forward-Backward SDEs Theory