ODE-DPS: ODE-based Diffusion Posterior Sampling for Inverse Problems in Partial Differential Equation

Enze Jiang,Jishen Peng,Zheng Ma,Xiong-Bin Yan

2024-04-21

Abstract:In recent years we have witnessed a growth in mathematics for deep learning, which has been used to solve inverse problems of partial differential equations (PDEs). However, most deep learning-based inversion methods either require paired data or necessitate retraining neural networks for modifications in the conditions of the inverse problem, significantly reducing the efficiency of inversion and limiting its applicability. To overcome this challenge, in this paper, leveraging the score-based generative diffusion model, we introduce a novel unsupervised inversion methodology tailored for solving inverse problems arising from PDEs. Our approach operates within the Bayesian inversion framework, treating the task of solving the posterior distribution as a conditional generation process achieved through solving a reverse-time stochastic differential equation. Furthermore, to enhance the accuracy of inversion results, we propose an ODE-based Diffusion Posterior Sampling inversion algorithm. The algorithm stems from the marginal probability density functions of two distinct forward generation processes that satisfy the same Fokker-Planck equation. Through a series of experiments involving various PDEs, we showcase the efficiency and robustness of our proposed method.

Numerical Analysis,Artificial Intelligence

What problem does this paper attempt to address?

### Problems Addressed by the Paper

This paper aims to address the inverse problems in partial differential equations (PDEs). Specifically, existing deep learning-based inverse methods often require paired data or retraining the neural network when conditions change, which significantly reduces the efficiency of solving inverse problems and limits their applicability. To tackle this challenge, the authors propose a novel unsupervised inverse method—an ordinary differential equation (ODE)-based diffusion posterior sampling algorithm (ODE-DPS) for solving inverse problems in PDEs.

#### Main Contributions:

1. **Novel Inverse Algorithm**:
- A method is proposed to learn the prior distribution of unknown parameters using partial prior data and a diffusion generative model.
- In the Bayesian inverse framework, the conditional generation process is achieved by solving the reverse-time stochastic differential equation.

2. **New Derivation Based on the Fokker-Planck Equation**:
- By using the Fokker-Planck equation satisfied by the marginal probability density function of the noise process in the generative model, a new reverse ordinary differential equation is derived.
- Numerical discretization results in the ODE-DPS inverse algorithm.

3. **Improved Adaptive Norm Residual Term**:
- Based on experimental observations, an adaptive norm residual term is proposed to replace the L2 norm data residual term to reduce errors near the boundaries.

4. **Efficiency and Accuracy**:
- Training can be conducted with only a small amount of prior data and without observational data (i.e., label data), improving inverse accuracy with minimal impact on efficiency.
- Experimental results show that this algorithm significantly improves inverse accuracy compared to traditional inverse methods.

Through these contributions, the paper aims to enhance the accuracy and efficiency of solving PDE inverse problems, especially in complex scenarios involving incomplete physical information and high-dimensional parameter spaces.