Title :
Methods of eliminating flat phenomenon during BP working
Author :
Wang, Kejun ; Jin, Hongzhang ; Li, Guobin
Author_Institution :
Dept. of Autom. Control, Harbin Eng. Univ., China
Abstract :
The measures for eliminating flat phenomenon are studied from several aspects. These aspects are activation function, the total input of neurons, the choice of activation functions, the adaptive learning of activate characteristic of neurons, and the choice of initial weights
Keywords :
adaptive systems; backpropagation; minimisation; neural nets; transfer functions; BP working; activate characteristic; activation function; activation functions; adaptive learning; backpropagation; flat phenomenon elimination; initial weights; neuron input; Automatic control; Error correction; Feedforward neural networks; Feeds; Gain measurement; Multi-layer neural network; Neural networks; Neurons; Tiles; Weight measurement;
Conference_Titel :
Intelligent Processing Systems, 1997. ICIPS '97. 1997 IEEE International Conference on
Conference_Location :
Beijing
Print_ISBN :
0-7803-4253-4
DOI :
10.1109/ICIPS.1997.672822
