Group Interventions on Deep Networks for Causal Discovery in Subsystems

2025-10-27

AI Summary

Most existing causal discovery methods operate at the pairwise variable level and thus fail to capture group-level causal relationships among subsystems (i.e., variable clusters). To address this limitation, we propose gCDMI—a novel deep neural network framework that jointly embeds subsystem-level interventions and model invariance testing for interpretable group-level causal structure discovery in nonlinear multivariate time series. Specifically, gCDMI performs controllable interventions on subsystems within the latent space and rigorously tests the invariance of downstream predictive models under such interventions to identify causally influential subsystem pairs. Extensive experiments on synthetic data, functional brain networks, and climate systems demonstrate that gCDMI significantly outperforms state-of-the-art pairwise causal discovery methods, achieving substantial improvements in both accuracy and robustness for group-level causal identification. This work establishes a new paradigm for hierarchical causal modeling in complex systems.

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: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Interdisciplinary science discovery with web data mining
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Causal discovery uncovers complex relationships between variables, enhancing predictions, decision-making, and insights into real-world systems, especially in nonlinear multivariate time series. However, most existing methods primarily focus on pairwise cause-effect relationships, overlooking interactions among groups of variables, i.e., subsystems and their collective causal influence. In this study, we introduce gCDMI, a novel multi-group causal discovery method that leverages group-level interventions on trained deep neural networks and employs model invariance testing to infer causal relationships. Our approach involves three key steps. First, we use deep learning to jointly model the structural relationships among groups of all time series. Second, we apply group-wise interventions to the trained model. Finally, we conduct model invariance testing to determine the presence of causal links among variable groups. We evaluate our method on simulated datasets, demonstrating its superior performance in identifying group-level causal relationships compared to existing methods. Additionally, we validate our approach on real-world datasets, including brain networks and climate ecosystems. Our results highlight that applying group-level interventions to deep learning models, combined with invariance testing, can effectively reveal complex causal structures, offering valuable insights for domains such as neuroscience and climate science.

Problem

• Identifies group-level causal relationships in multivariate time series
• Addresses limitations of pairwise methods by analyzing subsystem interactions
• Uses deep network interventions to reveal collective causal influences

Innovation

• Group-level interventions on trained deep networks
• Model invariance testing for causal inference
• Multi-group causal discovery in time series