Interpolation by neural network operators activated by ramp functions (Q2252489)

The theory of neural network (NN) operators has been introduced by \textit{G. A. Anastassiou} [J. Math. Anal. Appl. 212, No. 1, 237--262, Art. No. AY975494 (1997; Zbl 0899.68088)]. The problem of data interpolation and approximation by NN-operators is strictly related to the problem of the training of neural networks. In this paper, the author introduces a family of interpolation NN-operators activated by ramp functions as well as by sigmoidal functions generated by the central B-splines (see e.g. [\textit{D. Costarelli}, Ann. Mat. Pura Appl. (2013)], in press). Interpolation properties of the above operators are proved, together with a uniform approximation theorem with order, for continuous functions defined on bounded intervals. It is observed that a multivariate version of the theory developed in this paper, could be planned by adopting an approach that follows, for example in [\textit{G. A. Anastassiou}, Neural Netw. 24, No. 4, 378--386 (2011; Zbl 1228.65018)].