Publication Type
Conference Paper
Book Title
Proceedings of the NeurIPS 2023 AI for Science Workshop
Publication Date
Conference Name
NeurIPS 2023 AI for Science Workshop
Conference Location
New Orleans, Louisiana, United States of America
Conference Sponsor
The Neural Information Processing Systems Foundation
Conference Date
Abstract
In various scientific and engineering applications, there is typically an approximate model of the underlying complex system, even though it contains both aleatoric and epistemic uncertainties. In this paper, we present a principled method to incorporate these approximate models as physics priors in modeling, to prevent overfitting and enhancing the generalization capabilities of the trained models. Utilizing the structural risk minimization (SRM) inductive principle pioneered by Vapnik, this approach structures the physics priors into generalized regularizers. The experimental results demonstrate that our method achieves up to two orders of magnitude of improvement in testing accuracy.
