CausalDynamics: A large-scale benchmark for structural discovery of dynamical causal models

2025-05-22

AI Summary

Causal discovery in dynamic systems faces challenges including the difficulty of performing active interventions and limitations of existing benchmarks—such as determinism, low dimensionality, and weak nonlinearity. Method: This paper introduces DCM-Bench, the first scalable causal discovery benchmark specifically designed for complex dynamical systems. Built upon physically coupled ODE/SDE modeling frameworks, it supports high-dimensional, strongly nonlinear, stochastic dynamics, latent confounders, and time-delay effects. It encompasses thousands of synthetic systems and idealized climate models, generating realistic and ground-truth causal graphs. A standardized pipeline is proposed for causal graph synthesis, confounder injection, and time-delay modeling, accompanied by a plug-and-play Python toolkit and unified evaluation protocol. Contribution/Results: Extensive experiments systematically evaluate the robustness of mainstream causal discovery algorithms on complex dynamical systems, significantly improving reliability and generalizability in identifying causal structures—particularly for real-world applications such as climate modeling.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Knowledge Representation and Reasoning: Action, Change, and Causality

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Causal discovery for dynamical systems poses a major challenge in fields where active interventions are infeasible. Most methods used to investigate these systems and their associated benchmarks are tailored to deterministic, low-dimensional and weakly nonlinear time-series data. To address these limitations, we present CausalDynamics, a large-scale benchmark and extensible data generation framework to advance the structural discovery of dynamical causal models. Our benchmark consists of true causal graphs derived from thousands of coupled ordinary and stochastic differential equations as well as two idealized climate models. We perform a comprehensive evaluation of state-of-the-art causal discovery algorithms for graph reconstruction on systems with noisy, confounded, and lagged dynamics. CausalDynamics consists of a plug-and-play, build-your-own coupling workflow that enables the construction of a hierarchy of physical systems. We anticipate that our framework will facilitate the development of robust causal discovery algorithms that are broadly applicable across domains while addressing their unique challenges. We provide a user-friendly implementation and documentation on https://kausable.github.io/CausalDynamics.

Problem

• Challenges in causal discovery for dynamical systems without interventions
• Lack of benchmarks for noisy, high-dimensional nonlinear time-series data
• Need for robust causal discovery methods across diverse domains

Innovation

• Large-scale benchmark for dynamical causal models
• Extensible data generation framework with differential equations
• Plug-and-play workflow for building physical systems