Abstract:In this paper, we introduce an innovative approach for addressing Bayesian inverse problems through the utilization of physics-informed invertible neural networks (PI-INN). The PI-INN framework encompasses two sub-networks: an invertible neural network (INN) and a neural basis network (NB-Net). The primary role of the NB-Net lies in modeling the spatial basis functions characterizing the solution to the forward problem dictated by the underlying partial differential equation. Simultaneously, the INN is designed to partition the parameter vector linked to the input physical field into two distinct components: the expansion coefficients representing the forward problem solution and the Gaussian latent noise. If the forward mapping is precisely estimated, and the statistical independence between expansion coefficients and latent noise is well-maintained, the PI-INN offers a precise and efficient generative model for Bayesian inverse problems, yielding tractable posterior density estimates. As a particular physics-informed deep learning model, the primary training challenge for PI-INN centers on enforcing the independence constraint, which we tackle by introducing a novel independence loss based on estimated density. We support the efficacy and precision of the proposed PI-INN through a series of numerical experiments, including inverse kinematics, 1-dimensional and 2-dimensional diffusion equations, and seismic traveltime tomography. Specifically, our experimental results showcase the superior performance of the proposed independence loss in comparison to the commonly used but computationally demanding kernel-based maximum mean discrepancy loss.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to efficiently solve inverse problems within the Bayesian framework, especially by using Physics - Informed Invertible Neural Networks (PI - INN) to address this issue. Specifically, the paper aims to develop a method that can effectively generate samples of the Bayesian posterior distribution to overcome the problem of high computational cost of traditional methods such as Markov Chain Monte Carlo (MCMC) in high - dimensional situations. By combining physical information and deep - learning techniques, PI - INN can not only accurately predict the solutions of forward problems but also infer unknown parameters from observational data while ensuring the statistical independence of parameter estimates, thus providing an efficient and accurate solution to Bayesian inverse problems. ### Key points: 1. **Problem background**: - Inverse problems are common in many scientific and engineering fields, such as seismic tomography, material testing, and medical imaging. - Inverse problems are usually difficult to solve because observational data are often indirect and sparse, resulting in ill - posed problems. - Developing efficient computational models to obtain accurate and stable solutions has important practical and theoretical significance. 2. **Limitations of existing methods**: - Classical methods include regularization techniques and Bayesian inference. Regularization techniques determine point estimates by minimizing the error between the predicted output and the observational data and penalizing unwanted parameter characteristics. - Bayesian inference combines prior information and the uncertainty in observational data through Bayes' rule to construct the posterior distribution of parameters, but it has high computational complexity, especially in high - dimensional situations. 3. **Proposed method**: - **PI - INN framework**: It includes two sub - networks - Invertible Neural Network (INN) and Neural Basis Function Network (NB - Net). - **NB - Net**: It is used to model the spatial basis functions of the solutions of forward problems determined by partial differential equations (PDEs). - **INN**: It decomposes the parameter vector of the input physical field into two parts: the expansion coefficients representing the solutions of the forward problem and the Gaussian latent noise. - **Independence loss**: A new independence loss term is introduced to enforce the statistical independence between the expansion coefficients and the latent noise based on the estimated density. 4. **Experimental verification**: - The effectiveness and accuracy of PI - INN are verified through a series of numerical experiments, including inverse kinematics, one - dimensional and two - dimensional diffusion equations, and seismic travel - time tomography. - The experimental results show that the proposed independence loss term outperforms the commonly used kernel - based Maximum Mean Discrepancy (MMD) loss term in performance. ### Summary: This paper proposes a new method - Physics - Informed Invertible Neural Networks (PI - INN) for efficiently solving Bayesian inverse problems. By combining physical information and deep - learning techniques, PI - INN can provide accurate posterior distribution samples in high - dimensional situations, thus significantly reducing the computational cost and demonstrating its superior performance in multiple practical applications.

Efficient Bayesian inference using physics-informed invertible neural networks for inverse problems

Bayesian Inference and Deep Learning for Inverse Problems

Efficient Bayesian Physics Informed Neural Networks for inverse problems via Ensemble Kalman Inversion

B-PINNs: Bayesian physics-informed neural networks for forward and inverse PDE problems with noisy data

Surrogate modeling for Bayesian inverse problems based on physics-informed neural networks

Inverse Physics-Informed Neural Networks for transport models in porous materials

Data-Guided Physics-Informed Neural Networks for Solving Inverse Problems in Partial Differential Equations

A novel normalized reduced-order physics-informed neural network for solving inverse problems

Randomized Physics-Informed Neural Networks for Bayesian Data Assimilation

Learning functional priors and posteriors from data and physics

Bayesian Reasoning for Physics Informed Neural Networks

An Efficient Framework for Solving Forward and Inverse Problems of Nonlinear Partial Differential Equations Via Enhanced Physics-Informed Neural Network Based on Adaptive Learning

Transfer learning based physics-informed neural networks for solving inverse problems in engineering structures under different loading scenarios

Physics-informed graph neural Galerkin networks: A unified framework for solving PDE-governed forward and inverse problems

Deep fuzzy physics-informed neural networks for forward and inverse PDE problems

Physics-informed neural networks for solving forward and inverse problems in complex beam systems

Bayesian optimized physics-informed neural network for estimating wave propagation velocities

Solving forward and inverse PDE problems on unknown manifolds via physics-informed neural operators

Physics informed neural networks for an inverse problem in peridynamic models

Estimates on the generalization error of Physics Informed Neural Networks (PINNs) for approximating a class of inverse problems for PDEs

Maximum-likelihood Estimators in Physics-Informed Neural Networks for High-dimensional Inverse Problems