Physics-informed machine learning as a kernel method

Nathan Doumèche,Francis Bach,Gérard Biau,Claire Boyer

2024-06-19

Abstract:Physics-informed machine learning combines the expressiveness of data-based approaches with the interpretability of physical models. In this context, we consider a general regression problem where the empirical risk is regularized by a partial differential equation that quantifies the physical inconsistency. We prove that for linear differential priors, the problem can be formulated as a kernel regression task. Taking advantage of kernel theory, we derive convergence rates for the minimizer of the regularized risk and show that it converges at least at the Sobolev minimax rate. However, faster rates can be achieved, depending on the physical error. This principle is illustrated with a one-dimensional example, supporting the claim that regularizing the empirical risk with physical information can be beneficial to the statistical performance of estimators.

Artificial Intelligence,Statistics Theory

What problem does this paper attempt to address?

The paper primarily explores the issue of integrating physical knowledge into machine learning methods, particularly how to improve the performance of statistical learning through physical information. Specifically, the research focuses on regression problems within the framework of Physics-Informed Machine Learning (PIML), which combines data-driven methods with the interpretability of physical models.

The core contribution of the paper is to demonstrate that when the empirical risk is regularized by a partial differential equation representing physical inconsistency, it can be viewed as a kernel method. The authors prove that for linear differential priors, the problem can be transformed into a kernel regression task, and they derive an upper bound on the convergence rate of the estimator using kernel theory. Additionally, the paper discusses the impact of the physical regularization term on the convergence speed of the estimator, pointing out that if the target function strictly satisfies the given physical equation, the performance of the estimator will be significantly enhanced.

In short, the study aims to understand how physical information can improve the performance of estimators in statistical learning and provides theoretical guarantees. By analyzing the empirical risk minimization problem with physical regularization, the authors reveal that in certain cases, physical knowledge can lead to faster convergence rates, thereby supporting the beneficial role of physical information in machine learning.