Summary
International Symposium on Nonlinear Theory and Its Applications
2022
Session Number:A2L-D
Session:
Number:A2L-D-02
Variational Integrator for Hamiltonian Neural Networks
Yuhan Chen , Takashi Matsubara , Takaharu Yaguchi,
pp.25-28
Publication Date:12/12/2022
Online ISSN:2188-5079
DOI:10.34385/proc.71.A2L-D-02
PDF download (523.8KB)
Summary:
Hamiltonian neural networks are a type of neural networks for learning equations of motion that describe physical phenomena from given observed data. Such models should be used in physical simulations; however, it is known that when general-purpose numerical integrators are used for discretization, the energy conservation law and other laws of physics are destroyed. Structure-preserving numerical methods such as the variational integrator are effective to address this problem. We propose a variational integrator for Hamiltonian neural networks in this paper.