角分类前向神经网络研究  

Research on Forward Neural Network for Corner Classification

作　　者：张振亚[1] 陈恩红[2] 陈双平[1] 王进[2] 王煦法[2] 

机构地区：[1]中国科学技术大学电子工程与信息科学系,安徽合肥230027 [2]中国科学技术大学计算机系,安徽合肥230027

出　　处：《小型微型计算机系统》2005年第7期1215-1220,共6页Journal of Chinese Computer Systems

基　　金：教育部"面向二十一世纪教育振兴计划";中国博士后科学基金(2004036463)资助;国家自然科学基金资助项目(60005004)资助

摘　　要：角分类算法是一类快速分类算法,以其为学习算法的前向神经网络,在信息检索,特别是在线信息检索等领域有着重要的应用.通过对CC4学习算法的分析,揭示了泛化距离在角分类神经网络中的意义.针对文本数据的快速分类要求,提出了新的角分类网络TextCC.为解决数据的多类别判定问题,给出了新的角分类神经网络隐层与输出层之间连接矩阵的学习算法.实验表明,新的角分类神经网络隐层与输出层之间连接矩阵的学习算法有效,TextCC的分类精度教CC4的分类精度显著的提高.Corner classification (CC) is a kind of algorithms for instantly classification. The forward neural network trained by CC algorithm has important role on information retrieval, especially online information retrieval. The training algorithm of CC4 is analyzed in this paper and generalized distance is used to interpret the principle of training algorithm for CC4. A new network for CC, named as TextCC, is presented for classifying text object instantly. To give a solution for multi-corner judging, new training algorithm for the construction of connection matrix between hidden and output layer of forward neural network trained is presented. The experimental results show that the new training algorithm can work well on corner classification task and the precision of TextCC is improved markedly than CC4's.

关 键 词：前向神经网络 角分类 快速分类 泛化半径 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]