Romanian Journal of Information Technology and Automatic Control
Art. 08 – Vol. 25 – No. 1 – 2015
Liviana TUDOR
LTudor@upg-ploiesti.ro
Departamentul Informatică, Tehnologia Informaţiei, Matematică şi Fizică, Universitatea Petrol-Gaze din Ploieşti
Adrian MOISE
AMoise@upg-ploiesti.ro
Departamentul Automatică, Calculatoare şi Electronică, Universitatea Petrol-Gaze din Ploieşti
Abstract: This paper presents two methods to estimating errors in functions approximation, using the numerical and neural approach. The numerical method in function approximation uses Lagrange polynomials interpolation and describes the absolute interpolation error. The neural approach proves the property of a feed-forward artificial network to be a universal approximator and presents the neural errors propagation during the training process. The experimental results highlight the advantages of the intelligent neural systems in solving problems of function approximations..
Keywords: errors, function approximation, numerical methods, Lagrange polynomial, neural approximation, Backpropagation algorithm, feed-forward neural network.