Approximation by neural networks with scattered data

In this paper, the approximation capability of feed-forward neural networks (FNNs) formed as ∑k=1nckϕ(x-xk) is considered, where ϕ   is an even periodic continuous function and xkxk’s are scattered data. The best approximation error of trigonometric polynomials and the continuous modulus of the activation function ϕ are introduced to describe the approximation capability of FNNs.