Choice Function-Based Hyper-Heuristics for Causal Discovery under Linear Structural Equation Models.

Area of Science:

• Computational neuroscience
• Machine learning
• Causal inference

Background:

• Causal discovery, fundamental to cognition, relies on learning directed acyclic graphs (DAGs).
• Existing meta-heuristic algorithms for DAG learning often require domain-specific tuning and lack generalizability.
• Hyper-heuristics offer a promising alternative by combining and optimizing multiple heuristic algorithms.

Purpose of the Study:

• To propose a multi-population choice function hyper-heuristic for discovering causal relationships within DAGs.
• To integrate structural priors and expert knowledge with swarm intelligence for robust causal discovery.
• To enhance the generalization ability of DAG learning algorithms.

Main Methods:

• A multi-population choice function hyper-heuristic framework was developed.
• Partial v-structures were identified using partial correlation analysis to serve as structural priors.
• A linear structural equation model (SEM) was employed to guide the swarm intelligence approach.
• The search space was constrained through partial correlation analysis.

Main Results:

• The proposed hyper-heuristic method demonstrated effectiveness in causal discovery.
• Experimental results showed superior performance compared to existing state-of-the-art methods.
• The approach was validated on six standard benchmark networks.

Conclusions:

• The developed hyper-heuristic effectively combines swarm intelligence with structural priors for DAG learning.
• This method offers a robust solution for causal discovery, outperforming previous approaches.
• The findings highlight the potential of hyper-heuristics in advancing causal inference techniques.