一种自适应B样条网络学习算法

An Adaptive Learning Algorithm for B Spline Networks

本文提出Ｂ样条网络的一种自适应学习算法．在这种算法中，网络隐层Ｂ样条基函数的个数根据训练数据自动确定，而非零Ｂ样条基函数对应的内结点位置和连接权通过梯度下降法迭代调整．计算机模拟结果表明该算法比现有的Ｂ样条网络学习算法更加有效和实用．

In this paper an adaptive learning algorithm for B spline networks is presented,in which the number of B spline functions in the hidden layer is automatically determined from the information contained in training pairs and both weights and interior knot positions corresponding to the non zero B spline basis functions are iteratively adjusted by the gradient descent rule.Computer simulation results show that the proposed algorithm is more efficient and feasible than the existing learning algorithm used in B spline networks.