A PAC-Bayesian View of Generalisation for Physics-Informed Machine Learning

In three linesPAC-Bayesian framework for physics-informed machine learning (PIML) integrating partial differential equations. Provides high-probability generalisation guarantees with unbounded losses via multi-task perspective. Non-vacuous bounds validated on standard PDE benchmarks.

Summary generated by Claude — human-verified