Hamiltonian Graph Networks with ODE Integrators

Abstract

We introduce an approach for imposing physically informed inductive biases in learned simulation models. We combine graph networks with a differentiable ordinary differential equation integrator as a mechanism for predicting future states, and a Hamiltonian as an internal representation. We find that our approach outperforms baselines without these biases in terms of predictive accuracy, energy accuracy, and zero-shot generalization to time-step sizes and integrator orders not experienced during training. This advances the state-of-the-art of learned simulation, and in principle is applicable beyond physical domains.

References in corpus (5)

• Interaction Networks for Learning about Objects, Relations and Physics
• Hamiltonian Neural Networks
• A Compositional Object-Based Approach to Learning Physical Dynamics
• Latent ODEs for Irregularly-Sampled Time Series
• Differentiable Physics-informed Graph Networks

Cited by in corpus (44)

• Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges
• Learning to Simulate Complex Physics with Graph Networks
• Discovering Symbolic Models from Deep Learning with Inductive Biases
• Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data
• Symplectic Recurrent Neural Networks
• Lagrangian Neural Networks
• Graph Neural Ordinary Differential Equations
• Symplectic Neural Networks in Taylor Series Form for Hamiltonian Systems
• Port-Hamiltonian Neural Networks for Learning Explicit Time-Dependent Dynamical Systems
• Adaptable Hamiltonian neural networks
• gradSim: Differentiable simulation for system identification and visuomotor control
• LieTransformer: Equivariant self-attention for Lie Groups
• Differentiable Molecular Simulations for Control and Learning
• Neural Canonical Transformation with Symplectic Flows
• Continuous-in-Depth Neural Networks
• Simplifying Hamiltonian and Lagrangian Neural Networks via Explicit Constraints
• Deep Hamiltonian networks based on symplectic integrators
• Beltrami Flow and Neural Diffusion on Graphs
• MALI: A memory efficient and reverse accurate integrator for Neural ODEs
• Learning Physical Constraints with Neural Projections
• Machine learning structure preserving brackets for forecasting irreversible processes
• HamNet: Conformation-Guided Molecular Representation with Hamiltonian Neural Networks
• One-Shot Transfer Learning of Physics-Informed Neural Networks
• Unsupervised Learning of Lagrangian Dynamics from Images for Prediction and Control
• An Extensible Benchmark Suite for Learning to Simulate Physical Systems
• On Computational Poisson Geometry II: Numerical Methods
• Accurately Solving Physical Systems with Graph Learning
• Benchmarking Energy-Conserving Neural Networks for Learning Dynamics from Data
• Isometric Transformation Invariant and Equivariant Graph Convolutional Networks
• Simulating Continuum Mechanics with Multi-Scale Graph Neural Networks
• Continuous-Depth Neural Models for Dynamic Graph Prediction
• A simple equivariant machine learning method for dynamics based on scalars
• SympNets: Intrinsic structure-preserving symplectic networks for identifying Hamiltonian systems
• Group Equivariant Conditional Neural Processes
• Physics-enhanced Neural Networks in the Small Data Regime
• Time-Reversal Symmetric ODE Network
• Reconstruction of the interatomic forces from dynamic Scanning Transmission Electron Microscopy data
• Neural Symplectic Integrator with Hamiltonian Inductive Bias for the Gravitational $N$-body Problem
• Rodent: Relevance determination in differential equations
• Particle Track Reconstruction using Geometric Deep Learning
• Identifying Physical Law of Hamiltonian Systems via Meta-Learning
• Improving Simulations with Symmetry Control Neural Networks
• Nonseparable Symplectic Neural Networks
• Robust and Scalable SDE Learning: A Functional Perspective