Unsupervised selection and estimation of finite mixture models

Author

Figueiredo, Mário A T ; Jain, Anil K.

Author_Institution

Instituto de Telecomunicacoes, Inst. Superior Tecnico, Lisbon, Portugal

Volume

2

fYear

2000

fDate

2000

Firstpage

87

Abstract

We describe a method for fitting mixture models to multivariate data which performs component selection and does not require external initialization. The novelty of our approach includes: an MML-like (minimum message length) model selection criterion; inclusion of the criterion into the expectation-maximization (EM) algorithm (increasing its ability to escape from local maxima); an initialization strategy supported on the interpretation of EM as a self-annealing algorithm

Keywords

convergence; pattern clustering; probability; simulated annealing; statistical analysis; unsupervised learning; component selection; expectation-maximization algorithm; finite mixture models; initialization strategy; minimum message length-like model selection criterion; multivariate data; self-annealing algorithm; unsupervised estimation; unsupervised selection; Annealing; Bayesian methods; Clustering algorithms; Computer science; Integrated circuit modeling; Maximum likelihood estimation; Parameter estimation; Pattern recognition; Telecommunications; Unsupervised learning;

fLanguage

English

Publisher

ieee

Conference_Titel

Pattern Recognition, 2000. Proceedings. 15th International Conference on

Conference_Location

Barcelona

ISSN

1051-4651

Print_ISBN

0-7695-0750-6

Type

conf

DOI

10.1109/ICPR.2000.906023

Filename

906023