DocumentCode :
2595260
Title :
A combined method for improving backpropagation algorithm
Author :
Lee, Dong-Choon ; Kim, Byung-Joo ; Yun, Jeong-Mi ; Cha, Eui-Young
Author_Institution :
Dept. of Comput. Sci., Pusan Nat. Univ., South Korea
fYear :
1991
fDate :
13-16 Oct 1991
Firstpage :
1581
Abstract :
The authors present a new approach, called a combined method, for improving the backpropagation algorithm. They describe the standard backpropagation learning method and Newton´s method. Experiments on three problems are reported: exclusive-OR, the encoder and the parity problem. The learning speeds are compared by epochs and CPU time. The performance of the combined method and Newton´s method is compared. Experiments show that the combined method is faster than the standard backpropagation in epochs and more efficient than Newton´s method in spaces and computations
Keywords :
learning systems; neural nets; parallel algorithms; Newton´s method; backpropagation algorithm; combined method; encoder; exclusive-OR; learning algorithms; learning systems; neural nets; parity problem; Associative memory; Backpropagation algorithms; Computer science; Cost function; Filtering; Information processing; Learning systems; Neural networks; Neurons; Newton method;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Systems, Man, and Cybernetics, 1991. 'Decision Aiding for Complex Systems, Conference Proceedings., 1991 IEEE International Conference on
Conference_Location :
Charlottesville, VA
Print_ISBN :
0-7803-0233-8
Type :
conf
DOI :
10.1109/ICSMC.1991.169914
Filename :
169914
Link To Document :
بازگشت