A Fast Learning Strategy Using Pattern Selection for Feedforward Neural Networks

Szilárd Vajda 1 Yves Rangoni 1 Hubert Cecotti 1 Abdel Belaïd 1
1 READ - READ
LORIA - Laboratoire Lorrain de Recherche en Informatique et ses Applications
Abstract : Intelligent pattern selection is an active learning strategy where the classifiers select during training the most informative patterns. This paper investigates such a strategy where the informativeness of a pattern is measured as the approximation error produced by the classifier. The algorithm builds the training corpus starting from a small randomly chosen initial dataset and new patterns are added to the learning corpus based on their error sensitivity. The training dataset expansion is based on the selection of the most erroneous patterns. Our experimental results on MNIST 1 separated digit dataset show that only 3.26%of training data are sufficient for training purpose without decreasing the performance (98.36%) of the resulting neural classifier.
Keywords : Pattern selection incremental learning handwritten digit recognition neural networks support vector machines
Type de document :
Communication dans un congrès
Guy Lorette. Tenth International Workshop on Frontiers in Handwriting Recognition 2006 - IWFHR'10, Oct 2006, La Baule, France. Suvisoft, pp.6, 2006
Liste complète des métadonnées

https://hal.inria.fr/inria-00104833
Contributeur : Anne Jaigu <>
Soumis le : lundi 9 octobre 2006 - 14:59:40
Dernière modification le : mardi 25 octobre 2016 - 16:58:47
Document(s) archivé(s) le : mardi 6 avril 2010 - 19:02:39

Fichier

cr1109240398082.pdf

Identifiants

  • HAL Id : inria-00104833, version 1

Collections

INRIA | IWFHR10 | LORIA | UNIV-LORRAINE

Citation

Szilárd Vajda, Yves Rangoni, Hubert Cecotti, Abdel Belaïd. A Fast Learning Strategy Using Pattern Selection for Feedforward Neural Networks. Guy Lorette. Tenth International Workshop on Frontiers in Handwriting Recognition 2006 - IWFHR'10, Oct 2006, La Baule, France. Suvisoft, pp.6, 2006. <inria-00104833>

Exporter

BibTeX TEI DC DCterms EndNote

Partager

Métriques

Consultations de
la notice

190

Téléchargements du document

109

Contact


archive-ouverte@inria.fr