Abstract:Graph neural networks are often used to model interacting dynamical systems since they gracefully scale to systems with a varying and high number of agents. While there has been much progress made for deterministic interacting systems, modeling is much more challenging for stochastic systems in which one is interested in obtaining a predictive distribution over future trajectories. Existing methods are either computationally slow since they rely on Monte Carlo sampling or make simplifying assumptions such that the predictive distribution is unimodal. In this work, we present a deep state-space model which employs graph neural networks in order to model the underlying interacting dynamical system. The predictive distribution is multimodal and has the form of a Gaussian mixture model, where the moments of the Gaussian components can be computed via deterministic moment matching rules. Our moment matching scheme can be exploited for sample-free inference, leading to more efficient and stable training compared to Monte Carlo alternatives. Furthermore, we propose structured approximations to the covariance matrices of the Gaussian components in order to scale up to systems with many agents. We benchmark our novel framework on two challenging autonomous driving datasets. Both confirm the benefits of our method compared to state-of-the-art methods. We further demonstrate the usefulness of our individual contributions in a carefully designed ablation study and provide a detailed runtime analysis of our proposed covariance approximations. Finally, we empirically demonstrate the generalization ability of our method by evaluating its performance on unseen scenarios.

Expectation Propagation with Stochastic Kinetic Model in Complex Interaction Systems.

Optimizing Complex Interaction Dynamics in Critical Infrastructure with a Stochastic Kinetic Model.

Double-Chain Unscented Expectation Propagation for Inference in Stochastic Dynamical Models of Biological Processes.

Interacting Particle Systems on Networks: joint inference of the network and the interaction kernel

A Jump Stochastic Differential Equation Approach for Influence Prediction on Information Propagation Networks

Learning interpretable dynamics of stochastic complex systems from experimental data

Integrating Simulation and Signal Processing with Stochastic Social Kinetic Model.

Learning Theory for Inferring Interaction Kernels in Second-Order Interacting Agent Systems

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

Deep learning of stochastic contagion dynamics on complex networks

A jump stochastic differential equation approach for influence prediction on heterogenous networks

Computationally Efficient Modelling of Stochastic Spatio-Temporal Dynamics in Biomolecular Networks

Cheap and Deterministic Inference for Deep State-Space Models of Interacting Dynamical Systems

A Stochastic Model For Detecting Heterogeneous Link Communities In Complex Networks

Research of Complex Network Dynamics Evolution

Optimal Control of Complex Systems Through Variational Inference with a Discrete Event Decision Process.

Stochastic Prediction of Multi-Agent Interactions from Partial Observations

Accelerating inference for stochastic kinetic models

Community Structure Recovery and Interaction Probability Estimation for Gossip Opinion Dynamics

Dynamic Stochastic Block Model with Scale-Free Characteristic for Temporal Complex Networks

Non-intrusive model combination for learning dynamical systems