|
Activity Number:
|
164
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #305255 |
|
Title:
|
A Note on the Algorithmic Convergence of Posterior Simulation for Mixtures of Logistic Regression Model
|
|
Author(s):
|
Joy (Yang) Ge*+ and Wenxin Jiang
|
|
Companies:
|
Merck Research Laboratories and Northwestern University
|
|
Address:
|
351 N. Sumneytown Pike, North Wales, PA, 19454,
|
|
Keywords:
|
mixtures of experts ; Bayesian inference ; MCMC ; Metropolis-Hastings algorithm ; hybrid MCMC
|
|
Abstract:
|
Since their introduction, mixtures of experts (ME) and hierarchical mixtures of experts (HME) have become popular techniques for regression and classification. The underlying structure of the model is a mixture in which both the mixing coefficients and the mixing components are generalized linear models. Bayesian approaches can be used as a basis of inference and prediction. Computations can be performed using Markov Chain Monte Carlo (MCMC) methods. The Metropolis-Hastings algorithm is one such important MCMC method available for sampling from a posterior distribution. We will outline the sampling schemes to construct the simple/hybrid Metropolis-Hastings chains for ME with a nonstochastic/randome number of experts. Recent methods from Markov chain theory will be applied to both simple and hybrid algorithms and conditions for convergence of these algorithms will be established.
|
