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 > Information optimum vector quantization

Vendredi 17 mars 2006 à 11 heures

Information optimum vector quantization

Thomas Villmann (Université de Leipzig)

Abstract : Information optimum data processing is an important task in data analysis and data mining. We consider actual approaches for information optimal vector quantization. These approaches include methods which optimize information theoretic measures like Kullback-Leibler-divergence directly. Further, we show that for neural vector quantizer like self-organizing maps (SOMs) and neural gas (NG) information optimal data processing is possible by magnification control. Thereby, magnification is a property of the vector quantizer which is closely related to the description error by the law discovered by Zador. The effect of information control is demonstrated for several examples.

Dans la même rubrique :