Abstract:Schrödinger Bridge (SB) is an entropy-regularized optimal transport problem that has received increasing attention in deep generative modeling for its mathematical flexibility compared to the Scored-based Generative Model (SGM). However, it remains unclear whether the optimization principle of SB relates to the modern training of deep generative models, which often rely on constructing log-likelihood <a class="link-external link-http" href="http://objectives.This" rel="external noopener nofollow">this http URL</a> raises questions on the suitability of SB models as a principled alternative for generative applications. In this work, we present a novel computational framework for likelihood training of SB models grounded on Forward-Backward Stochastic Differential Equations Theory - a mathematical methodology appeared in stochastic optimal control that transforms the optimality condition of SB into a set of SDEs. Crucially, these SDEs can be used to construct the likelihood objectives for SB that, surprisingly, generalizes the ones for SGM as special cases. This leads to a new optimization principle that inherits the same SB optimality yet without losing applications of modern generative training techniques, and we show that the resulting training algorithm achieves comparable results on generating realistic images on MNIST, CelebA, and CIFAR10. Our code is available at <a class="link-external link-https" href="https://github.com/ghliu/SB-FBSDE" rel="external noopener nofollow">this https URL</a>.

Diffusion Schrödinger Bridge with Applications to Score-Based Generative Modeling

Simplified Diffusion Schrödinger Bridge

Diffusion Bridge Mixture Transports, Schrödinger Bridge Problems and Generative Modeling

Schödinger Bridge Type Diffusion Models as an Extension of Variational Autoencoders

Adversarial Schrödinger Bridge Matching

Diffusion Schrödinger Bridge Matching

Score-based Generative Modeling Through Backward Stochastic Differential Equations: Inversion and Generation

Reflected Schrödinger Bridge for Constrained Generative Modeling

Transport with Support: Data-Conditional Diffusion Bridges

Generative modeling for time series via Schr{ö}dinger bridge

Provably Convergent Schrödinger Bridge with Applications to Probabilistic Time Series Imputation

Likelihood Training of Schrödinger Bridge using Forward-Backward SDEs Theory

Variational Schrödinger Diffusion Models

Discrete Diffusion Schrödinger Bridge Matching for Graph Transformation

Deep Generative Learning via Schrödinger Bridge

Schrödinger Bridge Flow for Unpaired Data Translation

Soft-constrained Schrodinger Bridge: a Stochastic Control Approach

Score matching for sub-Riemannian bridge sampling

Conditional Simulation Using Diffusion Schrödinger Bridges

Go With the Flow: Fast Diffusion for Gaussian Mixture Models