Geometric Weights, Priors, and Their Applications

• Activity Number: 380
• Type: Invited
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Asociacion Mexicana de Estadistica
• Author(s): Ramsés H Mena*+
• Companies: Universidad Nacional Autónoma de México
• Keywords: Dirichlet Process ; Gibbs sampler ; Measure-valued process ; Nonparametric mixture model ; Nonparametric Regression

Abstract

Bayesian nonparametric techniques rely on suitable construction of nonparametric priors. The canonical example is without doubt the Dirichlet process, however in some situations different models are more suitable or preferred. Whereas most available alternatives to the Dirichlet process tend to generalized it to overcome certain prediction or fitting drawbacks, some of these issues might be rather solved with simpler models. Here we will review one of these simpler nonparametric priors, which can be seen as generated through a set of ordered weights within a species sampling model representation. We discuss various aspects of these random distributions as well as some of their applications in nonparametric mixtures, covariate or time dependent settings.