Abstract:This paper proposes a novel deep generative model, called BSDE-Gen, which combines the flexibility of backward stochastic differential equations (BSDEs) with the power of deep neural networks for generating high-dimensional complex target data, particularly in the field of image generation. The incorporation of stochasticity and uncertainty in the generative modeling process makes BSDE-Gen an effective and natural approach for generating high-dimensional data. The paper provides a theoretical framework for BSDE-Gen, describes its model architecture, presents the maximum mean discrepancy (MMD) loss function used for training, and reports experimental results.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to combine the flexibility of Backward Stochastic Differential Equations (BSDEs) with the powerful capabilities of deep neural networks when generating high - dimensional complex data (especially in the field of image generation), and propose a new deep - generation model BSDE - Gen. Specifically, the paper focuses on the following aspects: 1. **Generating high - dimensional complex data**: Existing generation models have limitations when dealing with high - dimensional complex data, especially in the field of image generation. The BSDE - Gen model aims to generate high - dimensional data more effectively by introducing randomness and uncertainty. 2. **Combining BSDEs and deep learning**: BSDEs can describe the reverse evolution of stochastic processes, while deep neural networks are good at modeling complex high - dimensional data. By combining these two, the paper proposes a novel framework for generating high - dimensional data. 3. **Theoretical basis and model architecture**: The paper provides the theoretical framework of the BSDE - Gen model and describes the model's architecture in detail, including the definitions of the forward process \(X_t\) and the backward process \(Y_t\), as well as the design of the control process \(Z_t\). 4. **Loss function and training method**: In order to train the BSDE - Gen model, the paper adopts the Maximum Mean Discrepancy (MMD) as the loss function and introduces two training strategies: decoder - only style and encoder - decoder style. 5. **Experimental verification**: The paper conducts experiments on two classic image datasets, MNIST and FashionMNIST, shows the effectiveness of the BSDE - Gen model, and discusses the limitations of the model and future research directions. Through these efforts, the paper hopes to provide a new tool for fields such as computer vision, biology, and drug discovery to simulate or generate complex systems with uncertain dynamics and incomplete information.

Deep Generative Modeling with Backward Stochastic Differential Equations

MBD-GAN: Model-based Image Deblurring with a Generative Adversarial Network.

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

A Method of Adaptive Hyperparameter Optimization for Deep Generative Models

Gradient boosting-based numerical methods for high-dimensional backward stochastic differential equations

A backward SDE method for uncertainty quantification in deep learning

Neural SDEs as Infinite-Dimensional GANs

Learning Deep Generative Models with Doubly Stochastic MCMC

Learning Deep Generative Models with Doubly Stochastic Gradient MCMC

Score-Based Generative Modeling through Stochastic Differential Equations

A forward differential deep learning-based algorithm for solving high-dimensional nonlinear backward stochastic differential equations

A Bayesian Non-parametric Approach to Generative Models: Integrating Variational Autoencoder and Generative Adversarial Networks using Wasserstein and Maximum Mean Discrepancy

A backward differential deep learning-based algorithm for solving high-dimensional nonlinear backward stochastic differential equations

Multi-dimensional Backward Stochastic Differential Equations of Diagonally Quadratic Generators with a Special Structure

Towards Robust and Stable Deep Learning Algorithms for Forward Backward Stochastic Differential Equations

Bridging Explicit and Implicit Deep Generative Models Via Neural Stein Estimators

Generative Modeling with Phase Stochastic Bridges

Backward stochastic differential equation approach to modeling of gene expression

Deep multi-step mixed algorithm for high dimensional non-linear PDEs and associated BSDEs

Deep Structured Generative Models.

Generative Modeling of Neural Dynamics via Latent Stochastic Differential Equations