DiffFlow: A Unified SDE Framework for Score-Based Diffusion Models and Generative Adversarial Networks

TL;DR

This paper introduces DiffFlow, a unified SDE framework that bridges score-based diffusion models and GANs, enabling flexible trade-offs between sample quality and sampling speed through adjustable dynamics.

Contribution

It presents a theoretical framework unifying SDMs and GANs via a novel SDE called DiffFlow, with proofs of optimality and new algorithms for improved generative modeling.

Findings

Unified DiffFlow SDE describes both SDMs and GANs dynamics.

Adjusting score weights transitions smoothly between models.

New algorithms achieve better trade-offs in sample quality and speed.

Abstract

Generative models can be categorized into two types: explicit generative models that define explicit density forms and allow exact likelihood inference, such as score-based diffusion models (SDMs) and normalizing flows; implicit generative models that directly learn a transformation from the prior to the data distribution, such as generative adversarial nets (GANs). While these two types of models have shown great success, they suffer from respective limitations that hinder them from achieving fast sampling and high sample quality simultaneously. In this paper, we propose a unified theoretic framework for SDMs and GANs. We shown that: i) the learning dynamics of both SDMs and GANs can be described as a novel SDE named Discriminator Denoising Diffusion Flow (DiffFlow) where the drift can be determined by some weighted combinations of scores of the real data and the generated data; ii) By adjusting the relative weights between different score terms, we can obtain a smooth transition between SDMs and GANs while the marginal distribution of the SDE remains invariant to the change of the weights; iii) we prove the asymptotic optimality and maximal likelihood training scheme of the DiffFlow dynamics; iv) under our unified theoretic framework, we introduce several instantiations of the DiffFLow that provide new algorithms beyond GANs and SDMs with exact likelihood inference and have potential to achieve flexible trade-off between high sample quality and fast sampling speed.