Learning Causes of Functional Dynamic Targets: Screening and Local Methods.

Area of Science:

• Causal inference
• Machine learning
• Time series analysis

Background:

• Identifying causal relationships in dynamic systems is crucial for understanding complex phenomena.
• Functional dynamic targets, which evolve over time based on multiple variables, present unique challenges for causal discovery.

Purpose of the Study:

• To develop robust methods for identifying direct and indirect causes of functional dynamic targets.
• To address the challenge of weakened correlations in long causal paths within dynamic systems.

Main Methods:

• A screening method is proposed to identify variables significantly correlated with the target.
• A combined approach of screening and structural learning is used to uncover causal structures.
• A local learning method is introduced to discover direct and indirect causes up to a specified distance, mitigating the distance effect.

Main Results:

• Theoretical analysis demonstrates the correctness of the proposed methods under regular assumptions.
• Experimental results using synthetic data confirm the effectiveness of the methods in learning target causes.

Conclusions:

• The developed screening and local learning methods provide an effective framework for causal discovery in functional dynamic targets.
• The approach successfully identifies both direct and indirect causal relationships, even in the presence of long causal paths.