Conditional Functional Graphical Models

Area of Science:

• Statistics
• Network Science
• Functional Data Analysis

Background:

• Multivariate functional data analysis is crucial in various fields.
• Existing graphical models often overlook subject-level heterogeneity influenced by external variables.
• Dynamic graphical modeling is essential for understanding time-varying network structures.

Purpose of the Study:

• To develop a conditional graphical model for multivariate random functions.
• To incorporate external variables to capture dynamic graph structures.
• To address subject-level heterogeneity in graphical modeling.

Main Methods:

• Introduced conditional precision and partial correlation operators.
• Extended precision and partial correlation matrices to conditional and functional settings.
• Developed estimators for characterizing conditional graphs.

Main Results:

• Established uniform convergence of estimators and consistency of estimated graphs.
• Demonstrated the ability to handle growing graph sizes and incomplete data.
• Validated the method using simulations and brain functional connectivity data.

Conclusions:

• The proposed conditional graphical model effectively handles multivariate functional data with external variables.
• The method provides a robust framework for dynamic network analysis, improving upon existing approaches.
• This technique has significant implications for fields analyzing complex, time-varying networks.