一个神经网络分类器的构造

Construction of a neural network classifier

提出了一个构造径向基函数神经网格分类器的有效方法 ,利用快速聚类和统计的方法确定网络中间层及中间层到输出层间的权值。把构造的分类器用于手写体数字的分类实验 ,取得了比较好的结果。

Radial basis function neural network providesa technique for approximating arbitrary nonlinear function mappings between nultidimensional spaces. Here a method is introduced to construct a RBFNN classfier by a fast clustering algorithm and the optimal weights are computed between the middle layer and output layer statistically.The method is proposed to construct an RBFNN classifier for handwritten digit recognition. The experiment showed that the method could construct a classifier quickly and the performance of the classifier was better than others.