Partenaires

CNRS
Logo tutelle
Logo tutelle
Logo tutelle


Rechercher

Sur ce site

Sur le Web du CNRS


Accueil du site > Séminaires > Probabilités Statistiques et réseaux de neurones > Prototype based fuzzy classification

Vendredi 10 mars 2006 à 11 heures

Prototype based fuzzy classification

Thomas Villmann (Université de Leipzig)

Abstract : Classification is an important field in data analysis. Prototype based methods like learning vector quantization (LVQ) and other provide an intuitive method which allows an understanding of the classification scheme, in contrast to multilayer perceptrons (MLPs) which work as a black box. Further, crisp classification some times in inadequate or impossble. Here fuzzy methods can help. We introduce extensions for supervised learning to the originally unsupervised prototype based neural vector quantizer self-organizing map (SOM) and neural gas (NG). Both approaches utilize neighbourhood cooperativness for improved convergence which is preserved in the supervised scheme, too. We demonstrate the approach for several examples including real world applications in bioinformatics.

Dans la même rubrique :