Title :
Reducing the herd effect in the backpropagation algorithm
Author :
Choong, Poh Lian ; Desilva, Christopher J S
Author_Institution :
Dept. of Electr. & Electron. Eng., Western Australia Univ., Nedlands, WA, Australia
Abstract :
The backpropagation algorithm is the most popular training algorithm for the multilayer perceptron, despite its slowness. Part of this slowness may be attributed to a phenomenon of the training process that has been called the herd effect. This paper describes a modification of the original backpropagation algorithm that is intended to reduce or eliminate the herd effect and reports on some performance comparisons.
Keywords :
backpropagation; convergence of numerical methods; multilayer perceptrons; backpropagation; dilution factor; herd effect; learning process; multilayer perceptron; neural nets; Artificial neural networks; Backpropagation algorithms; Convergence; Cows; Helium; Information processing; Intelligent systems; Mean square error methods; Multilayer perceptrons; Testing;
Conference_Titel :
Neural Networks, 1993. IJCNN '93-Nagoya. Proceedings of 1993 International Joint Conference on
Print_ISBN :
0-7803-1421-2
DOI :
10.1109/IJCNN.1993.713982
