Hamiltonian Neural Networks with Automatic Symmetry Detection

They also treat high dimensional data, in particular videos of mechanical systems. Modeling mechanical systems from first principles as Extending the HNN approach, Dierkes and Flaßkamp Hamiltonian or Lagrangian systems or using a Newton-Euler (2021) show how to learn a symmetry-preserving Hamiltonian, modeling approach has a long history. Recently, data-driven if the system symmetry is known a priori. Finzi et al. techniques have gained attention within this context to describe (2020) showed how neural networks can be made equivariant complex physical systems for which either no model and symmetric utilising convolutional layers with symmetric exists or existing models are too complicated to use in simulations.