Title :
Influence of Inputs in Modelling by Backpropagation Neural Networks
Author_Institution :
Higher Bus. Tech. Sch., Uzice
Abstract :
An influence of inputs in modelling processes by multilayer neural networks with backpropagation learning algorithm is given in the paper. Examination of input influence on an output error is performed by comparing the output error of network with and without a given input. Inputs significance, i.e. a measure of inputs influence on outputs, is represented by the final weights value. Influence of the distribution of inputs value on an approximation error is examined by determination of the output error for groups of inputs. The most important results of this analysis are the model optimization and reduction of the model error, which is applicable in practice
Keywords :
backpropagation; multilayer perceptrons; optimisation; powder metallurgy; production engineering computing; backpropagation learning algorithm; final weight value; input influence; multilayer neural networks; optimization; powder metallurgy; Artificial neural networks; Backpropagation algorithms; Biological neural networks; Biological system modeling; Desktop publishing; Digital audio players; ISO standards; Multi-layer neural network; Neural networks; Pressing; Backpropagation neural network; Input influence; Model error; Modelling;
Conference_Titel :
Neural Network Applications in Electrical Engineering, 2006. NEUREL 2006. 8th Seminar on
Conference_Location :
Belgrade, Serbia & Montenegro
Print_ISBN :
1-4244-0433-9
Electronic_ISBN :
1-4244-0433-9
DOI :
10.1109/NEUREL.2006.341210
