AMS2006-12: Dirichlet Process Mixtures of Beta Distributions, with Applications to Density and Intensity Estimation

Athanasios Kottas
12/31/2006 09:00 AM
Applied Mathematics & Statistics
We propose a class of Bayesian nonparametric mixture models with a Beta distribution providing the mixture kernel and a Dirichlet process prior assigned to the mixing distribution. Motivating applications include density estimation on bounded domains, and inference for non-homogeneous Poisson processes over time. We present the mixture model formulation, discuss prior specification, and develop a computational approach to posterior inference. The model is illustrated with two data sets.