Online Program Home
My Program

Abstract Details

Activity Number: 653 - Advances in Bayesian Nonparametric Modeling and Computation for Complex Data
Type: Topic Contributed
Date/Time: Thursday, August 2, 2018 : 10:30 AM to 12:20 PM
Sponsor: Section on Bayesian Statistical Science
Abstract #330481 Presentation
Title: Robustness and Nonparametric Bayesian Methods
Author(s): Steve MacEachern*
Companies: The Ohio State University

Nonparametric Bayesian models provide the flexibility needed to match the richness of modern data. They allow one to capture nonlinearities in regressions, changes in distributional form, variation in the degree of local flexibility, and complex dependence structures. The models are inherently infinite dimensional and so behave differently than low-dimensional parametric models. While the flexibility provided by these models allows them to uncover patterns in the data, this same flexibility makes them prone to overfitting-pursuing idiosyncrasies in a given data set that neither reflect underlying ``truth'' nor will be replicated in future data sets. The methods often exhibit excellent robustness at a global level but show a lack of robustness in some locales. A variety of strategies exist for improving the robustness of the methods, ranging from techniques for prior specification to constrain the flexibility of the models, to choice of the likelihood, to the mechanism for producing inference from the model. This talk will focus on mechanisms for enhancing robustness of nonparametric Bayesian methods.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2018 program