|
Activity Number:
|
225
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #303881 |
|
Title:
|
Bayesian Variable Selection in Generalized Gaussian Process Models
|
|
Author(s):
|
Terrance D. Savitsky*+ and Marina Vannucci and Naijun Sha
|
|
Companies:
|
Rice University and Rice University and The University of Texas at El Paso
|
|
Address:
|
Department of Statistics MS138, Houston, TX, 77251,
|
|
Keywords:
|
Bayesian variable selection ; Generalized functional model ; Gaussian process ; Generalized Gaussian process model ; Latent variables ; MCMC
|
|
Abstract:
|
In this paper we propose a Gaussian process modeling approach for the exponential dispersion family to facilitate variable selection in models with complex response surfaces. Our specific interest is in the analysis of data sets where predictors express an a priori unknown form and mix of linear and non-linear associations to the response. Starting with generalized linear models, we construct a new Gaussian process framework we label "generalized Gaussian process models" to incorporate a flexible nonparametric response surface function of predictors. Our framework employs Bayesian variable selection methods to both select predictors and identify the functional form of association to the response with no need to pre-specify the form, allowing for a parsimonious definition of the predictor space. Inference is done via MCMC techniques on simulated and benchmark data.
|
