Decomposing heterogeneous dynamical systems with graph neural networks

Cédric Allier,Magdalena C. Schneider,Michael Innerberger,Larissa Heinrich,John A. Bogovic,Stephan Saalfeld

2024-07-27

Abstract:Natural physical, chemical, and biological dynamical systems are often complex, with heterogeneous components interacting in diverse ways. We show that graph neural networks can be designed to jointly learn the interaction rules and the structure of the heterogeneity from data alone. The learned latent structure and dynamics can be used to virtually decompose the complex system which is necessary to parameterize and infer the underlying governing equations. We tested the approach with simulation experiments of moving particles and vector fields that interact with each other. While our current aim is to better understand and validate the approach with simulated data, we anticipate it to become a generally applicable tool to uncover the governing rules underlying complex dynamics observed in nature.

Machine Learning,Artificial Intelligence,Dynamical Systems

What problem does this paper attempt to address?

The paper aims to address the problem of how to use Graph Neural Networks (GNN) to learn and decompose heterogeneous dynamical systems from data. Specifically, the research objectives include: 1. **Designing Graph Neural Networks to jointly learn the interaction rules and heterogeneous structures of the system**: Starting solely from data, Graph Neural Networks can learn the intrinsic structure of complex systems and their dynamic behaviors. 2. **Virtually decomposing complex systems**: The learned latent structures and dynamics can be used to virtually decompose complex systems, which is crucial for parameterizing and inferring the underlying control equations. 3. **Testing the effectiveness of the method**: The authors validated the effectiveness of this method through simulation experiments involving moving particles and interacting vector fields. 4. **Becoming a generally applicable tool**: It is anticipated that this method will become a universally applicable tool for revealing the control rules behind the complex dynamics observed in nature. In short, the goal of this paper is to demonstrate that Graph Neural Networks can learn the intrinsic structure and dynamic behavior of complex, heterogeneous dynamical systems from data, and can virtually decompose these systems based on the learned knowledge, which helps in understanding and deriving the underlying control equations of these systems.

Decomposing heterogeneous dynamical systems with graph neural networks

Newton-Cotes Graph Neural Networks: on the Time Evolution of Dynamic Systems

Learning Coarse-Grained Dynamics on Graph

Learning interpretable dynamics of stochastic complex systems from experimental data

Learning graph geometry and topology using dynamical systems based message-passing

Towards complex dynamic physics system simulation with graph neural ordinary equations

When Graph Neural Networks Meet Dynamic Mode Decomposition

Efficient and scalable prediction of stochastic reaction-diffusion processes using graph neural networks

Neural Graph Simulator for Complex Systems

Equivariant Graph Hierarchy-Based Neural Networks

Coupled Graph ODE for Learning Interacting System Dynamics

A Graph Neural Network Approach to Dispersed Systems

Addressing Heterogeneity and Heterophily in Graphs: A Heterogeneous Heterophilic Spectral Graph Neural Network

Topological exploration of artificial neuronal network dynamics

Learning Networked Dynamical System Models with Weak Form and Graph Neural Networks

Learning Mesh-Based Simulation with Graph Networks

Learning Individual Interactions from Population Dynamics with Discrete-Event Simulation Model

Decomposed Linear Dynamical Systems (dLDS) for learning the latent components of neural dynamics

Heterogeneous Graph Propagation Network

Graph Neural Stochastic Differential Equations

Towards Complex Dynamic Physics System Simulation with Graph Neural ODEs