Sparse Bayesian infinite factor models.

Area of Science:

• Statistics and Machine Learning
• Computational Biology

Background:

• High-dimensional covariance matrices pose challenges for traditional statistical modeling.
• Bayesian latent factor models offer a flexible framework for dimensionality reduction.
• Existing methods often struggle with scalability and order dependence.

Purpose of the Study:

• To develop a novel sparse Bayesian latent factor model for high-dimensional data.
• To introduce a shrinkage prior for factor loadings enabling infinite factors.
• To enhance prediction and variable selection in complex datasets.

Main Methods:

• Proposed a multiplicative gamma process shrinkage prior on factor loadings.
• Developed an efficient, parameter-expanded Gibbs sampler for scalable computation.
• Introduced an adaptive Gibbs sampler for automatic truncation of the factor matrix.

Main Results:

• The proposed prior effectively shrinks loadings towards zero, allowing for infinite factors.
• The Gibbs sampler demonstrates computational efficiency and scalability with increasing dimensions.
• The method was successfully applied to predict survival times from gene expression data.

Conclusions:

• The novel Bayesian approach provides effective sparse modeling for high-dimensional covariance matrices.
• The developed methods offer efficient computation and improved prediction accuracy.
• This approach has significant potential for applications in bioinformatics and other high-dimensional fields.