How to speed up the training mechanism in a connectionist model
Identifieur interne : 004159 ( Crin/Curation ); précédent : 004158; suivant : 004160How to speed up the training mechanism in a connectionist model
Auteurs : Szilard Vajda ; Abdel BelaïdSource :
English descriptors
Abstract
In this paper a fast data driven learning-corpus building algorithm (FDDLCB) is proposed. The generic technique allows to build dynamically a representative and compact learning corpus for a connectionist model. The constructed dataset contains just a reduced number of patterns but sufficiently descriptive to characterize the different classes which should be separated. The method is based on a double least mean squares (LMS) error minimization mechanism trying to find the optimal boundaries of the different pattern classes. In the classical learning process the LMS is serving to minimize the error during the learning and this process is improved with a second one, as the new samples selection is also based on the idea to minimize the recognition error. Reinforcing the class boundaries where the recognition fails let us achieve a rapid and good generalization without any loss of occuracy. A modified version of the algorithm will be also presented. The experiments were performed on MNIST (Modified NIST) separated digit dataset. The encouraging result (98.51%) using just 1.85% of the available patterns from the original training dataset is comparable even with the state of the art techniques.
Links toward previous steps (curation, corpus...)
- to stream Crin, to step Corpus: Pour aller vers cette notice dans l'étape Curation :004159
Links to Exploration step
CRIN:vajda05bLe document en format XML
<record><TEI><teiHeader><fileDesc><titleStmt><title xml:lang="en" wicri:score="466">How to speed up the training mechanism in a connectionist model</title></titleStmt><publicationStmt><idno type="RBID">CRIN:vajda05b</idno><date when="2005" year="2005">2005</date><idno type="wicri:Area/Crin/Corpus">004159</idno><idno type="wicri:Area/Crin/Curation">004159</idno><idno type="wicri:explorRef" wicri:stream="Crin" wicri:step="Curation">004159</idno></publicationStmt><sourceDesc><biblStruct><analytic><title xml:lang="en">How to speed up the training mechanism in a connectionist model</title><author><name sortKey="Vajda, Szilard" sort="Vajda, Szilard" uniqKey="Vajda S" first="Szilard" last="Vajda">Szilard Vajda</name></author><author><name sortKey="Belaid, Abdel" sort="Belaid, Abdel" uniqKey="Belaid A" first="Abdel" last="Belaïd">Abdel Belaïd</name></author></analytic></biblStruct></sourceDesc></fileDesc><profileDesc><textClass><keywords scheme="KwdEn" xml:lang="en"><term>multi-layer perceptron</term><term>neural networks</term><term>pattern selection</term><term>separated digit recognition</term></keywords></textClass></profileDesc></teiHeader><front><div type="abstract" xml:lang="en" wicri:score="2984">In this paper a fast data driven learning-corpus building algorithm (FDDLCB) is proposed. The generic technique allows to build dynamically a representative and compact learning corpus for a connectionist model. The constructed dataset contains just a reduced number of patterns but sufficiently descriptive to characterize the different classes which should be separated. The method is based on a double least mean squares (LMS) error minimization mechanism trying to find the optimal boundaries of the different pattern classes. In the classical learning process the LMS is serving to minimize the error during the learning and this process is improved with a second one, as the new samples selection is also based on the idea to minimize the recognition error. Reinforcing the class boundaries where the recognition fails let us achieve a rapid and good generalization without any loss of occuracy. A modified version of the algorithm will be also presented. The experiments were performed on MNIST (Modified NIST) separated digit dataset. The encouraging result (98.51%) using just 1.85% of the available patterns from the original training dataset is comparable even with the state of the art techniques.</div></front></TEI><BibTex type="inproceedings"><ref>vajda05b</ref><crinnumber>A05-R-145</crinnumber><category>3</category><equipe>READ</equipe><author><e>Vajda, Szilard</e><e>Belaïd, Abdel</e></author><title>How to speed up the training mechanism in a connectionist model</title><booktitle>{IAPR - TC3 International Workshop on Neural Networks and Learning in Document Recognition - NNLDAR 2005, Seoul, Korea}</booktitle><year>2005</year><editor>Simone Marinai, Hiromichi Fujisawa</editor><pages>13--17</pages><month>Aug</month><note>A satellite workshop of ICDAR 2005</note><keywords><e>pattern selection</e><e>neural networks</e><e>multi-layer perceptron</e><e>separated digit recognition</e></keywords><abstract>In this paper a fast data driven learning-corpus building algorithm (FDDLCB) is proposed. The generic technique allows to build dynamically a representative and compact learning corpus for a connectionist model. The constructed dataset contains just a reduced number of patterns but sufficiently descriptive to characterize the different classes which should be separated. The method is based on a double least mean squares (LMS) error minimization mechanism trying to find the optimal boundaries of the different pattern classes. In the classical learning process the LMS is serving to minimize the error during the learning and this process is improved with a second one, as the new samples selection is also based on the idea to minimize the recognition error. Reinforcing the class boundaries where the recognition fails let us achieve a rapid and good generalization without any loss of occuracy. A modified version of the algorithm will be also presented. The experiments were performed on MNIST (Modified NIST) separated digit dataset. The encouraging result (98.51%) using just 1.85% of the available patterns from the original training dataset is comparable even with the state of the art techniques.</abstract></BibTex></record>Pour manipuler ce document sous Unix (Dilib)
EXPLOR_STEP=$WICRI_ROOT/Wicri/Lorraine/explor/InforLorV4/Data/Crin/Curation
HfdSelect -h $EXPLOR_STEP/biblio.hfd -nk 004159 | SxmlIndent | more
Ou
HfdSelect -h $EXPLOR_AREA/Data/Crin/Curation/biblio.hfd -nk 004159 | SxmlIndent | more
Pour mettre un lien sur cette page dans le réseau Wicri
{{Explor lien
|wiki= Wicri/Lorraine
|area= InforLorV4
|flux= Crin
|étape= Curation
|type= RBID
|clé= CRIN:vajda05b
|texte= How to speed up the training mechanism in a connectionist model
}}
| This area was generated with Dilib version V0.6.33. |  |