A Conditional Density Estimation Partition Model Using Logistic Gaussian Processes.

Area of Science:

• Statistics
• Machine Learning
• Computational Statistics

Background:

• Conditional density estimation (density regression) is crucial for understanding response variable distributions based on covariates.
• Existing methods may face challenges in efficiently learning complex partition structures.

Purpose of the Study:

• To propose a novel conditional density estimation method using a partition model framework.
• To leverage logistic Gaussian processes and Voronoi tessellations for enhanced density regression.

Main Methods:

• A partition model framework is employed, utilizing Voronoi tessellations learned via a reversible jump Markov chain Monte Carlo (MCMC) algorithm.
• Laplace approximation on latent variables of the logistic Gaussian process model enables efficient posterior distribution search of the tessellation.

Main Results:

• The proposed method demonstrates desirable consistency properties.
• Simulations and real-world applications show successful estimation of both the partition structure and the conditional distribution of the response variable.

Conclusions:

• The logistic Gaussian process-based partition model offers an effective approach for conditional density estimation.
• The method provides a computationally efficient and statistically sound framework for density regression.