Learning Energy Conserving Dynamics Efficiently with Hamiltonian Gaussian Processes

Magnus Ross, Markus Heinonen
3/3/2023
stat.MLcs.LG

No official code found. Best implementation: magnusross/hgp (3 stars, MIT, last commit 2023-03-14).

Abstract

Hamiltonian mechanics is one of the cornerstones of natural sciences. Recently there has been significant interest in learning Hamiltonian systems in a free-form way directly from trajectory data. Previous methods have tackled the problem of learning from many short, low-noise trajectories, but learning from a small number of long, noisy trajectories, whilst accounting for model uncertainty has not been addressed. In this work, we present a Gaussian process model for Hamiltonian systems with efficient decoupled parameterisation, and introduce an energy-conserving shooting method that allows robust inference from both short and long trajectories. We demonstrate the method's success in learning Hamiltonian systems in various data settings.

View on arXivView PDF

Code Implementations(1)

magnusross/hgpCommunity100%
MIT

Implementation of the HGP model from "Learning Energy Conserving Dynamics Efficiently with Hamiltonian Gaussian Processes"

30Python, Jupyter NotebookMar 2, 20233 years agoMIT

Similar Papers

Learning Dynamics from Input-Output Data with Hamiltonian Gaussian Processes

88% match2025

Learning Generalized Hamiltonian Dynamics with Stability from Noisy Trajectory Data

87% match2025

Gaussian Process Port-Hamiltonian Systems: Bayesian Learning with Physics Prior

85% match2023

Data-efficient Kernel Methods for Learning Hamiltonian Systems

84% match2025

Physics-Constrained Learning for PDE Systems with Uncertainty Quantified Port-Hamiltonian Models

83% match1 repo12024

Bayesian identification of nonseparable Hamiltonians with multiplicative noise using deep learning and reduced-order modeling

82% match2024

Cite this paper

@article{ross2023learning,
  title  = {Learning Energy Conserving Dynamics Efficiently with Hamiltonian Gaussian Processes},
  author = {Magnus Ross and Markus Heinonen},
  year   = {2023},
  eprint = {2303.01925},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2303.01925v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2303.01925.svg)](https://codeofpaper.com/paper/2303.01925)

Discussion