We discover that the prior knowledge embedded in the physics computation itself acts as an implicit regularization that greatly improves generalization of machine learning models for physics.
Please check out our recent paper:
The results serve as a proof of principle to rethink physics computation in the context of the new era of computing owing to achievements in automatic differentiation software, hardware and theories.
This work is a collaboration between Google Research (Li Li
@leeley18
, Stephan Hoyer
@shoyer
, Ruoxi Sun, Ekin Dogus Cubuk
@ekindogus
, Patrick Riley) and Burke group at UC Irvine (Ryan Pederson, Kieron Burke).