Multilayer feedforward networks are universal approximators
Neural Networks
A new pruning heuristic based on variance analysis of sensitivity information
IEEE Transactions on Neural Networks
Hi-index | 0.00 |
The neural network sensitivity analysis, involving neural network training and the calculation of its outputs derivative on inputs, was applied to select the least significant sensor in the multicomponent gas mixtures analysis system. The sensitivity analysis results, collected for various neural network structures were compared with the real significances of the sensors, determined experimentally. The question of the influence of the correlation of the input vector elements on the analysis results was also illustrated and discussed.