Two-Parameter Flows for Learning Population Dynamics of Physical Systems

In three linesNew method to learn dynamics of high-dimensional probability densities without labeled trajectories. Two-parameter flows learn sampling-time transports from base to marginals, then extract physics-time dynamics via regression on coupled synthetic trajectories. Scalable to high dimensions, handles rotational physics phenomena.

Summary generated by Claude — human-verified