A1489
Title: Variational Bayes and truncation approximations for enriched Dirichlet process mixtures
Authors: Somnath Bhadra - University of Florida (United States) [presenting]
Michael Joseph Daniels - University of Florida (United States)
Abstract: A common impediment in conducting inference for Bayesian nonparametric models is either the need for complex MCMC algorithms and/or computational run-time for large datasets. Solutions are proposed for enriched Dirichlet process mixtures (EDPM). A variational Bayes estimator is derived based on a previously developed truncation approximation for EDPMs. The variational Bayes estimator can be used in two ways: 1. to develop a more efficient truncation approximation; 2. as good initial values for a blocked Gibbs sampler based on this more efficient truncation approximation or for a polya urn sampler. The accuracy of this more efficient truncation approximation is derived, and it is demonstrated how this allows for simple implementation of EDPMs in standard software for Hamiltonian MC. The validity of the approximations is confirmed by simulations.
