Title: Beta-binomial stick breaking nonparametric prior
Authors: Ramses Mena - UNAM (Mexico) [presenting]
Abstract: A new class of stick-breaking nonparametric priors, termed Beta-Binomial process, is proposed. By allowing the underlying stick-breaking sequences to be dependent accordingly to a Beta-Binomial model, an appealing discrete random probability measure arises. Indeed, the resulting class contains the Dirichlet process and the Geometric process priors as particular cases. Tuning the chain's dependence parameter, controls the model's label switching adaptation for a given dataset and a given set of initial parameters. Some properties of the model are discussed and a density estimation algorithm proposed and tested with simulated datasets.