Higher generalization performance of artificial neural networks without reducing large databases
Michael Dipl.-Ing. Schneider, Fabrizio Calcagno, Michael Stieglitz, Ulrich Lehmann, Jörg Krone · Common Library Network (Der Gemeinsame Bibliotheksverbund) · 2010
This proceeding presents a possibility to enhance the accuracy of artificial neural networks.The database is divided in smaller parts and will be sorted.For each part an artificial neural network (ANN) will be trained.The overall result is calculated by the arithmetic mean value of all partial results from the trained ANNs.Also large databases, which usually can't be trained because of lack of resources, may be trained with this method.Consequently, the training time was accelerated and the total error was minimized, without reducing the data base and more ANNs.