A new algorithm of neural networks with B-spline weight functions

A new algorithm of neural networks with B-spline weight functions is proposed. The weights obtained after training are B-spline functions defined on the sets of input variables (input patterns), which can be used to extract some important information inherent in the problems. The new algorithm has high approximation accuracy and learning speed. The network´s architecture is very simple and the number of B-spline weight functions to be trained is independent of the number of patterns. Some examples are presented to illustrate good performance of the new algorithm.