Reflected Schrödinger Bridge for Constrained Generative Modeling

Wei Deng,Yu Chen,Nicole Tianjiao Yang,Hengrong Du,Qi Feng,Ricky T. Q. Chen

2024-01-06

Abstract:Diffusion models have become the go-to method for large-scale generative models in real-world applications. These applications often involve data distributions confined within bounded domains, typically requiring ad-hoc thresholding techniques for boundary enforcement. Reflected diffusion models (Lou23) aim to enhance generalizability by generating the data distribution through a backward process governed by reflected Brownian motion. However, reflected diffusion models may not easily adapt to diverse domains without the derivation of proper diffeomorphic mappings and do not guarantee optimal transport properties. To overcome these limitations, we introduce the Reflected Schrodinger Bridge algorithm: an entropy-regularized optimal transport approach tailored for generating data within diverse bounded domains. We derive elegant reflected forward-backward stochastic differential equations with Neumann and Robin boundary conditions, extend divergence-based likelihood training to bounded domains, and explore natural connections to entropic optimal transport for the study of approximate linear convergence - a valuable insight for practical training. Our algorithm yields robust generative modeling in diverse domains, and its scalability is demonstrated in real-world constrained generative modeling through standard image benchmarks.

Machine Learning

What problem does this paper attempt to address?

The paper addresses the optimization and boundary constraint issues of generative models in a bounded domain. Existing diffusion models often require threshold techniques to enforce boundaries when dealing with data distributions in practical applications. The paper introduces the Reflection-Schrodinger Bridge algorithm, which is a regularization method based on entropy for optimal transportation and applicable to various bounded domains for data generation. This algorithm extends divergence-based likelihood training through a novel reflective forward-backward stochastic differential equation, establishing a connection with entropy-optimal transportation. As a result, it enhances the robustness and scalability of generative models in different domains in both theoretical and practical aspects.

Reflected Schrödinger Bridge for Constrained Generative Modeling

Diffusion Bridge Mixture Transports, Schrödinger Bridge Problems and Generative Modeling

Latent Schr{ö}dinger Bridge Diffusion Model for Generative Learning

Discrete Diffusion Schrödinger Bridge Matching for Graph Transformation

Diffusion Schrödinger Bridge with Applications to Score-Based Generative Modeling

Latent Schrödinger Bridge Diffusion Model for Generative Learning

Schödinger Bridge Type Diffusion Models as an Extension of Variational Autoencoders

Deep Generative Learning Via Schrodinger Bridge

Deep Generative Learning via Schrödinger Bridge

One-step data-driven generative model via Schr\"odinger Bridge

Transport with Support: Data-Conditional Diffusion Bridges

Conditional Simulation Using Diffusion Schrödinger Bridges

Schrödinger Bridge Flow for Unpaired Data Translation

Stable generative modeling using Schrödinger bridges

Deep Generalized Schrödinger Bridges: From Image Generation to Solving Mean-Field Games

Generative Modeling with Phase Stochastic Bridges

Likelihood Training of Schrödinger Bridge using Forward-Backward SDEs Theory

Bridging Geometric States via Geometric Diffusion Bridge

Adversarial Schrödinger Bridge Matching

Generalized Schrödinger Bridge Matching