Training Emulators for Chaotic Dynamics We construct statistics-based losses for training more physically consistent emulators of high-dimensional chaos. Discovering Conservation Laws using Geometry We propose a direct geometric approach for discovering conservation laws using optimal transport and manifold learning. Discovering Sparse Interpretable Dynamics We discover sparse symbolic governing equations for dynamical systems using only partial state observations. Extracting Interpretable Physical Parameters Using a variational autoencoder, we identify and extract unknown physical parameters governing spatiotemporal dynamics.
