A systematic approach to the optimization of artificial neural networks

Author

Lin, T.Y. ; Ping, H.C. ; Hsu, T.H. ; Wang, L.C. ; Chen, C.C. ; Chen, C.F. ; Wu, C.S. ; Liu, T.C. ; Lin, C.L. ; Lin, Y.R. ; Chang, F.C.

Author_Institution

Dept. of Mechatron., Energy & Aerosp. Eng., Nat. Defense Univ., Taoyuan, Taiwan

fYear

2011

fDate

27-29 May 2011

Firstpage

76

Lastpage

79

Abstract

When designing the structure of an artificial neural network (ANN), it is very important to determine the architecture and parameters of the network such as number of units and layers. This paper uses the Taguchi method and Design of Experiment (DOE) methodology to optimize the network parameters. The users have to identify the application problems and choose a suitable ANN model. Then, the optimization problems including the design variables, cost function and constraints can be defined according to the network model. The Taguchi method is first applied to the problem for finding the more important factors. Then DOE methodology is performed for further analysis and forecast. Finally, the Multilayer Feed-forward Neural Network is used for an example.

Keywords

Taguchi methods; design of experiments; feedforward neural nets; optimisation; Taguchi method; artificial neural networks; cost function; design of experiment methodology; design variables; multilayer feedforward neural network; optimization problems; systematic approach; Artificial neural networks; Taguchi method; artificial neural network; design of experiment; optimization;

fLanguage

English

Publisher

ieee

Conference_Titel

Communication Software and Networks (ICCSN), 2011 IEEE 3rd International Conference on

Conference_Location

Xi´an

Print_ISBN

978-1-61284-485-5

Type

conf

DOI

10.1109/ICCSN.2011.6014853

Filename

6014853