径向基函数神经网络的一种有效的在线学习方法

AN EFFICIENT ON-LINE LEARNING METHOD FOR RADIAL BASIS FUNCTION NEURAL NETWORKS

本文提出了一种径向基函数神经网络的有效在线学习方法。该学习方法不仅能根据输入信息的增加而动态地分配网络资源,而且能有效回收网络的冗余资源。在学习过程中网络的参数可以自适应地序贯进行调整。文中详细论述了这种神经网络的学习准则、动态增减隐节点算法和参数调整算法。同时通过分析和实验说明网络具有较强的映射能力和预测性能。

This paper proposes an efficient on-line learning method for radial basis function (RBF) neural networks. The proposed learning method not only dynamically allocate the network resource in accordance with the increase of input Information, but also efficiently recycle the redundant resource of the network. During the learning process the parameters of the network can be sequentially adapted. The learning criterion, mechanism of increasing and decreasing resources and the parameter adjustment algorithm are elaborated. Meanwhile both the mapping approximation ability and predication performance of the network are analyzed in details.