Discovering Phase Space Structure in Learned Hamiltonian Systems
The Symplectic Phase Space GAN (SPS-GAN) is a generative model that learns the structure of complex dynamical systems, including their phase space, from raw data like trajectories and videos. It uses a port-Hamiltonian backbone and an adversarial setup to generate physically consistent dynamics. The model successfully infers the system's degrees of freedom and identifies underlying symmetries without needing prior knowledge of the configuration space.