Abstract:Physics-informed neural networks (PINNs) represent a significant advancement in scientific machine learning by integrating fundamental physical laws into their architecture through loss functions. PINNs have been successfully applied to solve various forward and inverse problems in partial differential equations (PDEs). However, a notable challenge can emerge during the early training stages when solving inverse problems. Specifically, data losses remain high while PDE residual losses are minimized rapidly, thereby exacerbating the imbalance between loss terms and impeding the overall efficiency of PINNs. To address this challenge, this study proposes a novel framework termed data-guided physics-informed neural networks (DG-PINNs). The DG-PINNs framework is structured into two distinct phases: a pre-training phase and a fine-tuning phase. In the pre-training phase, a loss function with only the data loss is minimized in a neural network. In the fine-tuning phase, a composite loss function, which consists of the data loss, PDE residual loss, and, if available, initial and boundary condition losses, is minimized in the same neural network. Notably, the pre-training phase ensures that the data loss is already at a low value before the fine-tuning phase commences. This approach enables the fine-tuning phase to converge to a minimal composite loss function with fewer iterations compared to existing PINNs. To validate the effectiveness, noise-robustness, and efficiency of DG-PINNs, extensive numerical investigations are conducted on inverse problems related to several classical PDEs, including the heat equation, wave equation, Euler--Bernoulli beam equation, and Navier--Stokes equation. The numerical results demonstrate that DG-PINNs can accurately solve these inverse problems and exhibit robustness against noise in training data.

On Stability and Regularization for Data-Driven Solution of Parabolic Inverse Source Problems

Solving the inverse potential problem in the parabolic equation by the deep neural networks method

A Dynamical Regularization Algorithm for Solving Inverse Source Problems of Elliptic Partial Differential Equations

Regularization of Inverse Problems by Neural Networks

Regularization Techniques for Estimating the Source in a Complete Parabolic Equation in $\mathbb{R}^n$

Well-posedness and Tikhonov regularization of an inverse source problem for a parabolic equation with an integral constraint

DRIP: Deep Regularizers for Inverse Problems

Stabilizing and Solving Inverse Problems using Data and Machine Learning

To be or not to be stable, that is the question: understanding neural networks for inverse problems

Stochastic Convergence of Regularized Solutions and Their Finite Element Approximations to Inverse Source Problems

Let data talk: data-regularized operator learning theory for inverse problems

Regularized solutions for some backward nonlinear parabolic equations with statistical data

Recovering the Source Term in Elliptic Equation Via Deep Learning: Method and Convergence Analysis

On an inverse problem in the parabolic equation arising from groundwater pollution problem

Stability of Data-Dependent Ridge-Regularization for Inverse Problems

Big in Japan: Regularizing networks for solving inverse problems

Regularization of the discrete source problem in the nonlinear diffusive-logistic equation

A data-driven adaptive regularization method and its applications

On the Asymptotical Regularization for Linear Inverse Problems in Presence of White Noise

Regularization with two differential operators and its application to inverse problems

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