反传神经网络的二分法使用  

Dichotomy Method with Back Propagation Neural Network in Model Classification

作　　者：李建中[1] 

机构地区：[1]包头师范学院学报编辑部,内蒙古包头014030

出　　处：《阴山学刊（自然科学版）》2005年第1期14-16,共3页Yinshan Academic Journal(Natural Science Edition)

摘　　要：本文中的二分法是指:把多分类问题逐级化为二分类问题;用结构简单的反传网络群代替结构复杂的单一网络;在用大样本对反传网络进行分类训练时分两步进行,第一步粗精度,使训练很快完成,然后根据网络的分类结果重新确定样本的类别,这样之后,第二步很快就可使训练达到高精度。Dichotomy in the paper is:multi classification project is gradually decomposed simple two classification projects;complex single BP network is took place of simple BP network group.When BP network is trained with great samples training would divide two steps.First step is with rough mean square error in order to training is completed rapidly. Primary classification of samples is anew identified base on experiment of BP network after first step.Precision of mean square error is quickly achieved subsequent fist step.

关 键 词：反传神经网络 分类 学习样本 训练网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]