Machine Learning with Physics Knowledge for Prediction: A Survey

Accepted at Transactions on Machine Learning Research (TMLR), 2025. The survey is rare in covering both academic methods and industrial deployment practices, plus an inventory of the physics-informed ML open-source ecosystem.

Best for: practitioners evaluating physics-informed ML for production use, especially teams needing a framework to pick between data-driven, physics-augmented, and meta-learning approaches.

Notable authors: Joe Watson, Jan Peters, Miles Cranmer et al. (TU Darmstadt + hessian.AI + Cambridge)