Common reducing subspace model and network alternation analysis.

Area of Science:

• Network analysis
• Statistical modeling
• Multivariate statistics

Background:

• Estimating covariance and precision matrices is crucial for understanding network structures in various fields, including neuroscience.
• Existing methods often struggle with high-dimensional data and lack a unified approach for both covariance and precision matrices.

Purpose of the Study:

• To propose a novel method for estimating covariance and precision matrices and their differences across multiple populations.
• To achieve substantial dimension reduction and efficient parameter estimation for network data.

Main Methods:

• A common reducing subspace model is proposed, extending the envelope model with a generalized sparsity principle.
• The method handles both covariance and precision matrices in a unified manner and accommodates matrix-valued data.
• Asymptotic analysis is used to quantify the efficiency gains of the proposed method.

Main Results:

• The proposed method demonstrates substantial dimension reduction and efficient parameter estimation.
• The approach offers a unified framework for covariance and precision matrix estimation, distinguishing itself from element-wise sparsity methods.
• Intensive simulations confirm the method's efficacy.

Conclusions:

• The common reducing subspace model provides an efficient and unified approach for network data analysis across multiple populations.
• This method offers advantages over existing techniques by handling both covariance and precision matrices and matrix-valued data.
• The approach is validated through simulations and applied to autism spectrum disorder data analysis.