Hamiltonian Neural Networks

TL;DR

This paper introduces Hamiltonian Neural Networks, which incorporate physical laws into neural network training to better learn conservation principles, resulting in faster training, improved generalization, and time-reversibility.

Contribution

The paper proposes a novel neural network architecture inspired by Hamiltonian mechanics that enforces conservation laws in an unsupervised manner.

Findings

Faster training compared to regular neural networks

Better generalization on physics-based problems

Model is perfectly reversible in time

Abstract

Even though neural networks enjoy widespread use, they still struggle to learn the basic laws of physics. How might we endow them with better inductive biases? In this paper, we draw inspiration from Hamiltonian mechanics to train models that learn and respect exact conservation laws in an unsupervised manner. We evaluate our models on problems where conservation of energy is important, including the two-body problem and pixel observations of a pendulum. Our model trains faster and generalizes better than a regular neural network. An interesting side effect is that our model is perfectly reversible in time.