Mastering high-dimensional dynamics with Hamiltonian neural networks

07/28/2020

∙

by   Scott T. Miller, et al.

∙

0

∙

share

We detail how incorporating physics into neural network design can significantly improve the learning and forecasting of dynamical systems, even nonlinear systems of many dimensions. A map building perspective elucidates the superiority of Hamiltonian neural networks over conventional neural networks. The results clarify the critical relation between data, dimension, and neural network learning performance.