|
Activity Number:
|
147
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, July 30, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #309463 |
|
Title:
|
Bayesian Kernel Regression with Feature Selection
|
|
Author(s):
|
Zhi Ouyang*+ and Merlise A. Clyde and Robert L. Wolpert
|
|
Companies:
|
Duke University and Duke University and Duke University
|
|
Address:
|
4225 Larchmont Rd, Durham, NC, 27707,
|
|
Keywords:
|
Bayesian kernel regression ; feature selection ; sparsity ; L\'{e}vy processes ; support vector machine ; relevance vector machine
|
|
Abstract:
|
We propose a general Bayesian framework to make feature selection using kernel regression. Unlike most kernel regression models in literature, our model allow kernel functions to have different shape parameters. Those shape parameters give us great flexibility to model different types of data. The approach uses L\'{e}vy priors to promote sparsity in both the kernel functions and predictor variables (or features); these priors act as regularizers for the likelihood function that reward good selected features. In large $p$ small $n$ problems, we keep the kernel centers within the observed data points, however, we could have more than one kernel centered at the same location. We relate this method to other work such as Support Vector Machines (SVM) and Relevance Vector Machine (RVM).
|
