Learning Neural Constitutive Laws

From Motion Observations for Generalizable PDE Dynamics