How to Improve the Generalization Ability of Multi-layer Neural Networks

0404606 - UIVT-O 20020032 RIV US eng C - Konferenční příspěvek (zahraniční konf.)
Šebesta, Václav
How to Improve the Generalization Ability of Multi-layer Neural Networks.
The 6th World Multi-Conference on Systemics, Cybernetics and Informatics. Proceedings. Vol. 6. Orlando: IIIS, 2002 - (Callaos, N.; Pisarchik, A.; Ueda, M.), s. 108-113. ISBN 980-07-8150-1.
[ISAS SCI 2002. World Multiconference on Systemics, Cybernetics and Informatics /6./. Orlando (US), 14.07.2002-18.07.2002]
Grant CEP: GA AV ČR IAA2030801; GA ČR GA102/02/0124
Výzkumný záměr: AV0Z1030915
Klíčová slova: neural networks topology * neural networks learning * generalization ability * prediction * classification * data mining
Kód oboru RIV: BA - Obecná matematika

The generalization ability of MNN will usually increase when the number of parameters, modified during the training process will decrease. We present an approach, based on the mathematical logic paradigms for the selection of significant input parameters, which are the most important from the point of view of output parameters. These parameters are later used for the training of multi-layer neural networks.
Trvalý link: http://hdl.handle.net/11104/0124849