DocumentCode
2664464
Title
A modified BP network based on second order momentum term
Author
Le, Liu ; Hongguo, Wang ; Baowei, Wang
Author_Institution
Sch. of Manage. & Econ., Shandong Normal Univ., Jinan
fYear
2008
fDate
16-18 July 2008
Firstpage
682
Lastpage
686
Abstract
On the base of BP network introducing second order momentum term, this paper proposed a modified method of setting second order momentum factor. This method exploited full advantage of adding first order momentum term, and set plus-minus of second order momentum factor according to different situations of BP network on the error surface during training process. At the end of the paper, simulation experiments were made on BP network applying the new method by solving three benchmark problems: XOR, 3-Bit Parity and function approximation. The results of the experiments show that the new BP network has faster training speed and better training effect.
Keywords
backpropagation; function approximation; neural nets; 3-bit parity; XOR; error surface; function approximation; modified BP network; second order momentum factor; second order momentum term; training process; Electronic mail; Function approximation; Information science; Benchmark problems; Error surface; Gradient descent; Second order momentum term;
fLanguage
English
Publisher
ieee
Conference_Titel
Control Conference, 2008. CCC 2008. 27th Chinese
Conference_Location
Kunming
Print_ISBN
978-7-900719-70-6
Electronic_ISBN
978-7-900719-70-6
Type
conf
DOI
10.1109/CHICC.2008.4605423
Filename
4605423
Link To Document