Learning Markov Blankets From Multiple Interventional Data Sets.

Area of Science:

• Machine Learning
• Causal Inference
• Statistical Learning

Background:

• Markov blankets (MBs) are crucial for causal Bayesian network structure learning, feature selection, and domain adaptation.
• Existing MB discovery methods primarily focus on single observational datasets, limiting their application to interventional data.

Purpose of the Study:

• To develop a method for learning Markov blankets from multiple interventional datasets.
• To address challenges of unknown intervention variables and non-identical data distributions in MB discovery.

Main Methods:

• Theoretical analysis of conditions for correct MB discovery and cause identification from interventional data.
• Proposal of a novel algorithm for MB learning from multiple interventional datasets.
• Validation using benchmark Bayesian networks and real-world datasets.

Main Results:

• Established theoretical conditions for accurate Markov blanket discovery under interventions.
• Developed and validated a new algorithm effective for learning MBs from multiple interventional datasets.
• Demonstrated the algorithm's effectiveness and efficiency in experimental settings.

Conclusions:

• The proposed algorithm and theoretical framework advance Markov blanket discovery from interventional data.
• This work provides the first theoretical analysis and algorithmic solution for MB discovery across multiple interventional datasets.
• The findings have significant implications for causal discovery and robust feature selection in machine learning.