Learning to Simulate Complex Physics with Graph Networks

Abstract

Here we present a machine learning framework and model implementation that can learn to simulate a wide variety of challenging physical domains, involving fluids, rigid solids, and deformable materials interacting with one another. Our framework---which we term "Graph Network-based Simulators" (GNS)---represents the state of a physical system with particles, expressed as nodes in a graph, and computes dynamics via learned message-passing. Our results show that our model can generalize from single-timestep predictions with thousands of particles during training, to different initial conditions, thousands of timesteps, and at least an order of magnitude more particles at test time. Our model was robust to hyperparameter choices across various evaluation metrics: the main determinants of long-term performance were the number of message-passing steps, and mitigating the accumulation of error by corrupting the training data with noise. Our GNS framework advances the state-of-the-art in learned physical simulation, and holds promise for solving a wide range of complex forward and inverse problems.

arXiv commentAccepted at ICML 2020

References in corpus (15)

• Adam: A Method for Stochastic Optimization
• Semi-Supervised Classification with Graph Convolutional Networks
• Neural Message Passing for Quantum Chemistry
• Interaction Networks for Learning about Objects, Relations and Physics
• Dynamic Graph Convolutional Networks
• Graph networks as learnable physics engines for inference and control
• Learning to Predict the Cosmological Structure Formation
• Learning Particle Dynamics for Manipulating Rigid Bodies, Deformable Objects, and Fluids
• Flexible Neural Representation for Physics Prediction
• Hamiltonian Graph Networks with ODE Integrators
• DiffTaichi: Differentiable Programming for Physical Simulation
• Sinkhorn Distances: Lightspeed Computation of Optimal Transportation Distances
• Relational Forward Models for Multi-Agent Learning
• Scale MLPerf-0.6 models on Google TPU-v3 Pods
• Stochastic Prediction of Multi-Agent Interactions from Partial Observations

Cited by in corpus (100)

• Machine learning accelerated computational fluid dynamics
• A Generalization of Transformer Networks to Graphs
• Physics-informed graph neural Galerkin networks: A unified framework for solving PDE-governed forward and inverse problems
• Discovering Symbolic Models from Deep Learning with Inductive Biases
• Benchmarking Graph Neural Networks
• Object-Centric Learning with Slot Attention
• Transformers for Modeling Physical Systems
• Solver-in-the-Loop: Learning from Differentiable Physics to Interact with Iterative PDE-Solvers
• Graph neural networks for laminar flow prediction around random 2D shapes
• Weak baselines and reporting biases lead to overoptimism in machine learning for fluid-related partial differential equations
• DiffCloth: Differentiable Cloth Simulation with Dry Frictional Contact
• Multi-resolution partial differential equations preserved learning framework for spatiotemporal dynamics
• Graph Convolutional Neural Networks for Body Force Prediction
• Graph Neural Networks Accelerated Molecular Dynamics
• Graph neural networks for simulating crack coalescence and propagation in brittle materials
• Three-dimensional deep learning-based reduced order model for unsteady flow dynamics with variable Reynolds number
• DiffAqua: A Differentiable Computational Design Pipeline for Soft Underwater Swimmers with Shape Interpolation
• DIG: A Turnkey Library for Diving into Graph Deep Learning Research
• Thermodynamics-informed graph neural networks
• MAgNET: A Graph U-Net Architecture for Mesh-Based Simulations
• Port-Hamiltonian Neural Networks for Learning Explicit Time-Dependent Dynamical Systems
• Learning Physical Graph Representations from Visual Scenes
• Robust Prediction of Force Chains in Jammed Solids using Graph Neural Networks
• Learning the Travelling Salesperson Problem Requires Rethinking Generalization
• Fluid Simulation on Neural Flow Maps
• Soft Metamaterials: Adaptation and Intelligence
• Hybrid quantum physics-informed neural networks for simulating computational fluid dynamics in complex shapes
• Finite Volume Graph Network(FVGN): Predicting unsteady incompressible fluid dynamics with finite volume informed neural network
• Graph Neural Network contextual embedding for Deep Learning on Tabular Data
• On the Use of Neural Networks for Full Waveform Inversion
• Directed Acyclic Graph Neural Networks
• Spherical Message Passing for 3D Graph Networks
• ForceNet: A Graph Neural Network for Large-Scale Quantum Calculations
• Photoacoustic image synthesis with generative adversarial networks
• Applying Convolutional Neural Networks to Data on Unstructured Meshes with Space-Filling Curves
• Manipulation of Granular Materials by Learning Particle Interactions
• Visual Grounding of Learned Physical Models
• Multi-agent Trajectory Prediction with Fuzzy Query Attention
• Pointer Graph Networks
• Hypersolvers: Toward Fast Continuous-Depth Models
• Multi-scale Time-stepping of Partial Differential Equations with Transformers
• GraphGLOW: Universal and Generalizable Structure Learning for Graph Neural Networks
• Graph Coarsening with Neural Networks
• 3D-OES: Viewpoint-Invariant Object-Factorized Environment Simulators
• Learning to simulate and design for structural engineering
• Augmenting Differentiable Simulators with Neural Networks to Close the Sim2Real Gap
• MPMNet: A Data-Driven MPM Framework for Dynamic Fluid-Solid Interaction
• Scalable variational Monte Carlo with graph neural ansatz
• Predicting Rigid Body Dynamics using Dual Quaternion Recurrent Neural Networks with Quaternion Attention
• On the Unreasonable Effectiveness of Feature propagation in Learning on Graphs with Missing Node Features
• Graph Neural Network Based Surrogate Model of Physics Simulations for Geometry Design
• Finite-difference-informed graph network for solving steady-state incompressible flows on block-structured grids
• Efficient Graph Deep Learning in TensorFlow with tf_geometric
• From Local Structures to Size Generalization in Graph Neural Networks
• AG2U -- Autonomous Grading Under Uncertainties
• Graph Neural Networks with Parallel Neighborhood Aggregations for Graph Classification
• Direct Prediction of Steady-State Flow Fields in Meshed Domain with Graph Networks
• Forward Prediction for Physical Reasoning
• DrivAer Transformer: A high-precision and fast prediction method for vehicle aerodynamic drag coefficient based on the DrivAerNet++ dataset
• Noether Networks: Meta-Learning Useful Conserved Quantities
• Scalar Field Prediction on Meshes Using Interpolated Multi-Resolution Convolutional Neural Networks
• Predicting Change, Not States: An Alternate Framework for Neural PDE Surrogates
• Understanding complex crowd dynamics with generative neural simulators
• PhysGNN: A Physics-Driven Graph Neural Network Based Model for Predicting Soft Tissue Deformation in Image-Guided Neurosurgery
• SENSORIMOTOR GRAPH: Action-Conditioned Graph Neural Network for Learning Robotic Soft Hand Dynamics
• Learning the dynamics of a one-dimensional plasma model with graph neural networks
• Conditionally Parameterized, Discretization-Aware Neural Networks for Mesh-Based Modeling of Physical Systems
• Trajectory Prediction using Equivariant Continuous Convolution
• Frame invariance and scalability of neural operators for partial differential equations
• Slot Contrastive Networks: A Contrastive Approach for Representing Objects
• On Computational Poisson Geometry II: Numerical Methods
• Learning Long-term Visual Dynamics with Region Proposal Interaction Networks
• Multi-Grid Graph Neural Networks with Self-Attention for Computational Mechanics
• Accurately Solving Physical Systems with Graph Learning
• Isometric Transformation Invariant and Equivariant Graph Convolutional Networks
• Integrating Physics and Topology in Neural Networks for Learning Rigid Body Dynamics
• A Model-Constrained Tangent Slope Learning Approach for Dynamical Systems
• Learning Continuous Network Emerging Dynamics from Scarce Observations via Data-Adaptive Stochastic Processes
• Local Off-Grid Weather Forecasting with Multi-Modal Earth Observation Data
• $\mathcal{H}$-HIGNN: A Scalable Graph Neural Network Framework with Hierarchical Matrix Acceleration for Simulation of Large-Scale Particulate Suspensions
• Deep learning of material transport in complex neurite networks
• Interactive High-Resolution Simulation of Granular Material
• Predicting Real-time Scientific Experiments Using Transformer models and Reinforcement Learning
• Fea2Fea: Exploring Structural Feature Correlations via Graph Neural Networks
• In-Depth Investigation of Phase Transition Phenomena in Network Models Derived from Lattice Models
• Dynamic Relational Inference in Multi-Agent Trajectories
• Particle Track Reconstruction using Geometric Deep Learning
• Reasoning-Modulated Representations
• Identifying Physical Law of Hamiltonian Systems via Meta-Learning
• A Deep Emulator for Secondary Motion of 3D Characters
• Physical Reasoning Using Dynamics-Aware Models
• Discovering Latent Representations of Relations for Interacting Systems
• Multigrid Solver With Super-Resolved Interpolation
• Symmetry-driven graph neural networks
• The Role of Isomorphism Classes in Multi-Relational Datasets
• SeaPearl: A Constraint Programming Solver guided by Reinforcement Learning
• Invertible Surrogate Models: Joint surrogate modelling and reconstruction of Laser-Wakefield Acceleration by invertible neural networks
• Data-Driven Continuum Dynamics via Transport-Teleport Duality
• Learning Mechanically Driven Emergent Behavior with Message Passing Neural Networks
• Spectral Transform Forms Scalable Transformer