Title :
Selection of minimum training data for generalization and online training by multilayer neural networks
Author :
Hara, Kazuyuki ; Nakayama, Kenji
Author_Institution :
Graduate Sch. of Nat. Sci. Technol., Kanazawa Univ., Japan
Abstract :
A training data reduction method for a multilayer neural network (MLNN) is proposed in this paper. This method reduce the data by selecting the minimum number of training data that guarantee generality of the MLNN. For this purpose, two methods are used: 1) a pairing method which selects the training data by finding the nearest data of the different classes, and data along the class boundary in data space can be selected; and 2) a training method which uses a semi-optimum MLNN in a training process. Since the MLNN classify data based on the distance from the network boundary, the selected data can be located close to the class boundary. So, if the semi-optimum MLNN did not select data from class boundary, pairing method can select them. The methods proposed can be applied to both off-line training and online training. The methods are investigated through computer simulation
Keywords :
backpropagation; feedforward neural nets; generalisation (artificial intelligence); pattern classification; real-time systems; backpropagation; generalization; multilayer neural networks; network boundary; online training; pairing method; pattern classification; supervised learning; training data selection; Computer architecture; Computer simulation; Data mining; Educational institutions; Multi-layer neural network; Network address translation; Neural networks; Nonhomogeneous media; Supervised learning; Training data;
Conference_Titel :
Neural Networks, 1996., IEEE International Conference on
Conference_Location :
Washington, DC
Print_ISBN :
0-7803-3210-5
DOI :
10.1109/ICNN.1996.548932
