具有渐进局部学习特性的多色Voronoi分类器设计

Design of Multicolor Voronoi Classifier with Gradually Local Learning Ablility

本文提出了一种多色Voronoi分类器MCVC.MCVC在学习样本上有好的边界推广性随着样本数量的增加MCVC的分类面可以逼近任意的分类函数.MCVC具有好的局部特性,对新加样本的训练只影响其周围的局部性态,不会对全局产生大的影响,可以克服神经网络方法对样本的过学习问题.实验表明MCVC对于线性和非线性分类问题都具有最优分类面。

A novel MultiColor Voronoi Classifier(MCVC)is proposed, which can be applied to linear and nonlinear classification problems. MCVC has sound ability to expend classification plane between samples. With increment of samples, it can be shown that the classification plane of MCVC can close to any classification function. MCVC has very good local ability too. When new learning sample is added, only local classification planed is modified and the whole classification characteristics are not modified greatly. So MCVC can solve the overfitting problem of neural network. Experiments show that MCVC is feasible to linear classification and nonlinear classification problems.