Physics-informed neural networks (PINN)

Raissi-Perdikaris-Karniadakis 2019 PINN. NN trained to satisfy PDE in residual + boundary-conditions in loss. Solves forward + inverse PDE problems. Bridges ML + PDE.