Event Details
Comparisons Among Information-Based Clustering Criteria, a Novel Modification Thereof, and the Monte Carlo Markov Chain Method
Presenter: Xiaoli Lu
Supervisor:
Date: Mon, August 27, 2001
Time: 14:00:00 - 15:00:00
Place: EOW 430
ABSTRACT
ABSTRACT:
One of the important elements in blind clustering is to estimate the cluster number. We study and analyze those approximation methods based on K-L distance to build a simplified expression for information criteria in practice. Simulations and expreiements using Gaussian mixture model are presented to prove the validity and rightness of Simplified Information Criterion. The comparison between S-BIC , S-CAIC and Birth-death Markov chain Monte Carlo (BDMCMC), a popular classification way from Bayesian approach to several typical data are shown and coresponding results give us an understanding in those cluster number estimating methods.