A Double Regression Method for Graphical Modeling of High-dimensional Nonlinear and Non-Gaussian Data.

Area of Science:

• Statistics
• Machine Learning
• Data Science

Background:

• Graphical models are essential for inferring conditional independence in large datasets.
• Existing methods primarily address Gaussian or linearly dependent data, limiting their application.
• High-dimensional, nonlinear, and non-Gaussian data present significant challenges for current graphical modeling techniques.

Purpose of the Study:

• To develop a robust method for learning graphical models in high-dimensional, nonlinear, and non-Gaussian settings.
• To address the limitations of existing graphical modeling approaches that assume linearity or Gaussian distributions.
• To establish theoretical consistency guarantees for the proposed method under mild conditions.

Main Methods:

• A novel double regression approach is proposed for graphical model learning.
• The method employs a series of nonparametric conditional independence tests.
• A double regression procedure, utilizing sure independence screening or sparse deep neural networks, reduces the conditioning set for tests.

Main Results:

• The proposed double regression method demonstrates consistency under mild conditions.
• Numerical results confirm the method's effectiveness with high-dimensional, nonlinear, and non-Gaussian data.
• The approach successfully infers conditional independence relationships in complex data structures.

Conclusions:

• The double regression method offers a powerful new tool for graphical model learning in challenging data environments.
• This work extends the applicability of graphical models to a broader range of real-world datasets.
• The proposed technique provides a statistically sound and computationally viable solution for complex dependency structures.