Variable Selection for Nonparametric Gaussian Process Priors: Models and Computational Strategies.

Area of Science:

• Statistics
• Machine Learning
• Computational Statistics

Background:

• Gaussian process models are powerful tools for regression and classification.
• Existing models often struggle with complex data structures and unknown predictor relationships.
• There is a need for flexible, unified models applicable to various data types.

Purpose of the Study:

• To present a unified treatment of Gaussian process models.
• To extend these models to handle data from the exponential dispersion family and survival data.
• To address the analysis of data with predictors exhibiting unknown nonlinear associations.

Main Methods:

• Incorporating Gaussian processes within a generalized linear model framework.
• Developing nonparametric regression models where the covariance matrix depends on predictors.
• Exploring alternative covariance formulations and mixture priors for variable selection.
• Utilizing Markov Chain Monte Carlo (MCMC) strategies for posterior inference.

Main Results:

• A flexible class of nonparametric regression models for continuous, categorical, and count responses.
• Models capable of accounting for survival outcomes.
• A computationally efficient and practical framework for variable selection using mixture priors.
• Demonstrated performance on simulated and benchmark datasets.

Conclusions:

• The proposed unified Gaussian process framework offers significant flexibility and broad applicability.
• The methods provide a robust approach for analyzing complex data with unknown predictor associations.
• The developed variable selection and inference strategies are computationally efficient and practical.