A0966
Title: Bayesian nonparametric models with BART components
Authors: Maria Kalli - Kings College London (United Kingdom) [presenting]
Jim Griffin - University College London (United Kingdom)
Abstract: Bayesian additive regression tree, BART, models have emerged as an important method for Bayesian nonlinear regression. However, there is little work on fully nonparametric versions of BART. Bayesian nonparametric models are described, built using BART elements, which can be easily implemented using existing BART R packages. A general density regression method is developed using normalized latent measure random factor models to build mixtures where the components are heteroscedastic BARTs. The performance of the method is evaluated on both simulated and real data.
