Extended Hamiltonian Learning on Riemannian Manifolds: Theoretical Aspects