User menu

Accès à distance ? S'identifier sur le proxy UCLouvain

Flexible and Robust Bayesian Classification by Finite Mixture Models

Primary tabs

Archambeau, Cédric [UCL] Vrins, Frédéric [UCL] Verleysen, Michel [UCL]

The regularized Mahalanobis distance is proposed in the framework of nite mixture models to avoid commonly faced numerical difficulties encountered with EM. Its principle is applied to Gaussian and Student-t mixtures, resulting in reliable density estimates, the model complexity being kept low. Besides, the regularized models are robust to various noise types. Finally, it is shown that the quality of the associated Bayesian classication is near optimal on Ripley's synthetic data set.

Document type Communication à un colloque (Conference Paper) – Présentation orale avec comité de sélection
Access type Accès restreint
Publication date 2004
Language Anglais
Conference "ESANN 2004, European Symposium on Artificial Neural Networks", Bruges (Belgium) (du 28/04/2004 au 30/04/2004)
Peer reviewed yes
Host document "Proceedings of ESANN 2004, European Symposium on Artificial Neural Networks"- p. 75-80
Affiliation UCL - FSA/ELEC - Département d'électricité
Links
Bibliographic reference Archambeau, Cédric ; Vrins, Frédéric ; Verleysen, Michel. Flexible and Robust Bayesian Classification by Finite Mixture Models.ESANN 2004, European Symposium on Artificial Neural Networks (Bruges (Belgium), du 28/04/2004 au 30/04/2004). In: Proceedings of ESANN 2004, European Symposium on Artificial Neural Networks, 2004, p. 75-80
Permanent URL http://hdl.handle.net/2078.1/110973