“…Backpropagation (BP) algorithm is a kind of neural computing algorithm, but its drawbacks have been made widespread, such as local minima, slow convergence, or dependent on initialized values in the training procedure and so on. To overcome the drawbacks of BP algorithm and some early algorithms, the new kinds of weight functions have been proposed in [2,3,4,5,6] for training neural networks, in which the weights are called weight functions, for example, the weights are cubic spline functions [3,4,5], B-spline weight functions [5], or improved B-spline weight functions [6] and so on. Although weight function theories and methods have been used in training neural networks in recent years and have many advantages, the trained weight functions are polynomials, which may be difficult for some rational problems.…”