An Expectation Conditional Maximization approach for Gaussian graphical models.

Area of Science:

• Computational statistics
• Machine learning
• Statistical modeling

Background:

• Bayesian graphical models are valuable for understanding complex variable dependencies, especially with prior information.
• High-dimensional data presents computational challenges for traditional Bayesian stochastic search methods.
• Existing methods struggle with the vast search space of possible graphs in high-dimensional scenarios.

Purpose of the Study:

• To propose a computationally efficient and deterministic alternative to existing Bayesian graphical model estimation methods.
• To extend the Expectation-Maximization (EM) algorithm for Bayesian variable selection to graphical model estimation.
• To enable the estimation of Gaussian and Gaussian copula graphical models in high-dimensional settings.

Main Methods:

• Developed a novel Expectation Conditional Maximization (ECM) algorithm.
• Extended the EM algorithm framework for Bayesian graphical model estimation.
• Applied the ECM algorithm to Gaussian and Gaussian copula graphical models.

Main Results:

• The proposed ECM algorithm provides a computationally feasible deterministic alternative.
• Demonstrated fast posterior exploration capabilities of the ECM approach.
• Showcased the ability of the ECM algorithm to incorporate multiple sources of prior information.

Conclusions:

• The ECM algorithm offers an efficient method for estimating Bayesian graphical models, particularly in high-dimensional data.
• This approach overcomes the computational limitations of traditional stochastic search methods.
• The ECM algorithm facilitates the integration of external prior information into graphical model estimation.