Title :
Convergence properties and stationary points of the two-layer backpropagation algorithm used for nonlinear function modeling
Author :
Ibnkahla, Mohamed
Author_Institution :
Electr. & Comput. Eng. Dept., Queen´s Univ., Kingston, Ont.
Abstract :
The paper presents a statistical analysis of the two-layer backpropagation algorithm. The network is applied to a function approximation problem (for modeling a traveling wave tube transfer function which is used in satellite communications) with noisy input ant output measurements. The network mean squared error surface is expressed as functions of the input and output measurement noise variances, the input signal variance, and the network weights. The paper proposes recursions which predict the mean weight behavior during the learning process. Computer simulations show good agreement between theory and experimental results
Keywords :
backpropagation; convergence of numerical methods; feedforward neural nets; function approximation; least mean squares methods; statistical analysis; backpropagation; convergence; function approximation; learning process; mean squared error surface; measurement noise; multilayer neural networks; network weights; signal variance; statistical analysis; Backpropagation algorithms; Convergence; Function approximation; Neural networks; Neurons; Noise measurement; Satellite communication; Signal processing algorithms; Statistical analysis; Transfer functions;
Conference_Titel :
Neural Networks, 2001. Proceedings. IJCNN '01. International Joint Conference on
Conference_Location :
Washington, DC
Print_ISBN :
0-7803-7044-9
DOI :
10.1109/IJCNN.2001.939097
