Learning smooth dynamics via Kinematic Interpolants

Inferring dynamics from real-world experimental data is inherently ambiguous, as many plausible dynamical systems can explain the same positional observations at measurement times.