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 > Recursive Self-Organizing Maps for Structures

Vendredi 18 mars 2005, à 10h

Recursive Self-Organizing Maps for Structures

Barbara Hammer (Technical University of Clausthal, Allemagne)

Résumé : The Self-Organizing Map (SOM) and alternatives like Neural Gas (NG) constitute powerful and universal tools for clustering, data-mining, and visualization. However, in their original form, they have been proposed for vectorial data, and problems arise if more complex data such as time series or structures are considered. In recent years, several different extensions of SOM to recursive structures have been proposed based on a variety of different schemes of time and context integration, the temporal Kohonen map (TKM) being one of the earliest approaches, the SOM for structured data and recursive SOM being two recent ones. In this talk, we propose a general framework for different context and time integration schemes and we discuss the benefits of the specific choices with respect to real-life applications. We add a further model which can be interpreted as a more powerful realization of the original TKM and demonstrate its benefit within a task for speaker identification. Finally, we have a look at principled theoretical issues. These include the question of cost functions for learning, the representational issues, induced metrics, and the connection to classical mechanisms for processing time series such as finite automata.

Voir en ligne : le site de Barbara Hammer

Dans la même rubrique :