DocumentCode :
1902712
Title :
Sensitivity analysis for input vector in multilayer feedforward neural networks
Author :
Fu, Li ; Chen, Tinghuai
Author_Institution :
Chongqing Univ., China
fYear :
1993
fDate :
1993
Firstpage :
215
Abstract :
The derivative matrix, or the Jacobian matrix, of the output vector with respect to the input vector is obtained for multilayer feedforward neural networks (MFNNs). This matrix represents the sensitivity to small perturbations in the input of an MFNN. The expression for the matrix describes the performance of the MFNN, such as the generalization capabilities, as well as error-correcting properties. Analysis shows how these aspects of performance are affected by the weight matrices, the sigmoid functions, and the number of layers and nodes of the network. Suggestions are made for the design of MFNNs with good generalization and error-correction
Keywords :
error correction; feedforward neural nets; generalisation (artificial intelligence); matrix algebra; sensitivity analysis; Jacobian matrix; derivative matrix; error-correcting properties; generalization capabilities; input vector; multilayer feedforward neural networks; sigmoid functions; small perturbations; weight matrices; Computer networks; Feedforward neural networks; Intelligent networks; Jacobian matrices; Mathematics; Multi-layer neural network; Neural networks; Performance analysis; Sensitivity analysis; Systems engineering and theory;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Neural Networks, 1993., IEEE International Conference on
Conference_Location :
San Francisco, CA
Print_ISBN :
0-7803-0999-5
Type :
conf
DOI :
10.1109/ICNN.1993.298559
Filename :
298559
Link To Document :
بازگشت