Reduction of Gaussian mixture models by maximum similarity

Scott and Szewczyk developed an iterative method to simplify (reduce the order of) a Gaussian mixture model by merging the two most similar components. Since the comparison of all pairs of components may not be feasible, they propose to consider only nearly adjacent components, with no guarantee tha...

Full description

Saved in:
Bibliographic Details
Published in:Journal of nonparametric statistics 2010-08, Vol.22 (6), p.703-709
Main Author: Harmse, Jørgen E.
Format: Article
Language:English
Subjects:
60-04
62-07
Algorithms
Comparative analysis
Estimating techniques
Gaussian
Gaussian mixture models
Iterative methods
kernel density estimation
machine learning and neural networks
Mathematical models
Merging
model simplification
Nonparametric statistics
Normal distribution
Reduction
Similarity
Citations: Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Scott and Szewczyk developed an iterative method to simplify (reduce the order of) a Gaussian mixture model by merging the two most similar components. Since the comparison of all pairs of components may not be feasible, they propose to consider only nearly adjacent components, with no guarantee that they find the most similar. I give a method to find the most similar pair of components without comparing all pairs, and I propose an extension to higher dimensions.
ISSN:1048-5252
1029-0311
DOI:10.1080/10485250903377293