Title :
Learning algorithm for neural network with alternating process
Author :
Liu, Wei-Guo ; Yong, Lin ; Huang, Yong-xuan
Author_Institution :
Syst. Eng. Inst., Xi´´an Jiaotong Univ., China
Abstract :
This paper is concerned with improving the learning efficiency and stability of neural networks. An alternating algorithm is presented for the training process. The method is divided into two stages. First, the gradient method is applied, and then a bisection technique is used. The convergence is proved for the proposed method, so that a stable distribution of weights can be reached. Compared with the standard gradient method, the oscillating divergent phenomenon is avoided.
Keywords :
convergence; feedforward neural nets; gradient methods; learning (artificial intelligence); stability; alternating algorithm; bisection technique; convergence; feedforward neural network; gradient method; learning efficiency; oscillating divergent phenomenon avoidance; stability; stable weight distribution; training process; Artificial neural networks; Convergence; Educational institutions; Electronic mail; Feedforward neural networks; Feeds; Gradient methods; Neural networks; Signal processing algorithms; Systems engineering and theory;
Conference_Titel :
Machine Learning and Cybernetics, 2002. Proceedings. 2002 International Conference on
Print_ISBN :
0-7803-7508-4
DOI :
10.1109/ICMLC.2002.1174482
