Mastering high-dimensional dynamics with Hamiltonian neural networks

2020/07/28 by Scott T. Miller, Miller, Scott T., John F. Lindner +7
Computer Science · Physics and Astronomy · #Chaotic Dynamics (nlin.CD) #Computational Physics and Python Applications #FOS: Computer and information sciences #FOS: Physical sciences #Model Reduction and Neural Networks #Neural Networks and Applications #Neural and Evolutionary Computing (cs.NE)

paper · pdf · doi:10.48550/arxiv.2008.04214

openalex publication_date 2020/07/28 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

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.

Related

• Learning Neural Hamiltonian Dynamics: A Methodological Overview
• Hamiltonian Generative Networks
• Forecasting Hamiltonian dynamics without canonical coordinates
• Learning Hamiltonian Dynamics at Scale: A Differential-Geometric Approach
• Applications of Machine Learning to Modelling and Analysing Dynamical Systems