Title :
The causes for premature saturation with backpropagation training
Author :
Vitela, Javier E. ; Reifman, Jaques
Author_Institution :
Inst. de Ciencias Nucl., Univ. Nacional Autonoma de Mexico, Mexico City, Mexico
fDate :
27 Jun-2 Jul 1994
Abstract :
The mechanism that causes the output nodes of feedforward multilayer networks mapped with sigmoid functions to prematurely saturate during backpropagation training is described. The necessary conditions for the occurrence of this undesirable phenomenon are also presented. Simulation results demonstrate the adequacy of the presented necessary conditions
Keywords :
backpropagation; feedforward neural nets; multilayer perceptrons; backpropagation training; feedforward multilayer networks; necessary conditions; output nodes; premature saturation; sigmoid functions; Algorithm design and analysis; Backpropagation algorithms; Convergence; Inductors; Iterative algorithms; Joining processes; Laboratories; Nonhomogeneous media;
Conference_Titel :
Neural Networks, 1994. IEEE World Congress on Computational Intelligence., 1994 IEEE International Conference on
Conference_Location :
Orlando, FL
Print_ISBN :
0-7803-1901-X
DOI :
10.1109/ICNN.1994.374499
