Neural Approximation of Empirical Functions |
| J. Roj
Institute of Measurement Science, Electronics and Control at the Silesian University of Technology, Akademicka 10, 44-100 Gliwice, Poland |
| Full Text PDF |
| The paper presents the results of simulation studies of selected neural network structures used for non-linear function approximation based on a limited accuracy data. There was performed the analysis of the interdependence of the network structure and the size of the set of learning patterns. The approximation inaccuracy was expressed by the uncertainty interval width. The approximation properties of the neural method were compared with those of the piece-wise linear and polynomial: "cubic" and "spline" methods. |
DOI: 10.12693/APhysPolA.124.554 PACS numbers: 84.35.+i, 07.05.Mh, 06.20.-f, 06.20.F-, 06.20.fb, 02.60.Ed |