DocumentCode :
1752174
Title :
Deficiency in the current trend of training of neural network systems, suggestions and solutions
Author :
Tien, Dapeng ; Nobar, Peter
Author_Institution :
Sch. of Inf. Technol., Charles Sturt Univ., Australia
Volume :
1
fYear :
2001
fDate :
2001
Firstpage :
43
Abstract :
Although artificial neural networks have been experimented extensively, many users pay little or no attention to the internal strum of such systems. As a result, inefficient algorithms were commonly used and much result was obtained on an ad hoc basic. The conventional training methods are not suitable for networks with large number of neurons. Furthermore, the learning rate constant can easily affect the convergence and the rate of convergence. In this paper, a number of non-linear optimisation algorithms have been proposed for training neural network systems with large number of neurons. Because of the strong mathematical background of these algorithms they can be used to train difficult neural networks with a single layer. The results have shown that the speed of the networks can be increased by several hundred times
Keywords :
convergence; learning (artificial intelligence); neural nets; optimisation; transfer functions; Kolmogorov´s theorem; activation functions; artificial neural networks; artificial neuron; biological neuron; convergence; learning rate; nonlinear optimisation; synaptic strength; training methods; Approximation algorithms; Artificial neural networks; Biological neural networks; Biological system modeling; Biology computing; Computer networks; Humans; Intelligent networks; Neural networks; Neurons;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
TENCON 2001. Proceedings of IEEE Region 10 International Conference on Electrical and Electronic Technology
Print_ISBN :
0-7803-7101-1
Type :
conf
DOI :
10.1109/TENCON.2001.949548
Filename :
949548
Link To Document :
بازگشت