首页 | 本学科首页   官方微博 | 高级检索  
     

具有双向检测机制的ART2神经元网络
引用本文:钟旭,陈德钊,陈亚秋,罗建宏.具有双向检测机制的ART2神经元网络[J].浙江大学学报(自然科学版 ),2004,38(12):1540-1544.
作者姓名:钟旭  陈德钊  陈亚秋  罗建宏
作者单位:浙江大学化工系,浙江大学化工系,浙江大学化工系,浙江大学化工系 浙江杭州310027,浙江杭州310027,浙江杭州310027,浙江杭州310027
摘    要:为了解决传统的自适应共振理论(T-ART2)网络和增设最初模板的ART2(OP-ART2)网络对模式类渐变过程的不敏感性而导致的识别性能较差的缺陷,提出具有双向检测机制的ART2(BM-ART2)网络.它增设两个门限,对输入模式的聚类进行双向检测,并采用更能反映模式本身特性的比较层输出计算相似度.对一组等腰三角形模式的聚类测试表明,BM-ART2网络能识别模式类的渐变过程,杜绝或减少了模式超出与交混等异常现象,其聚类性能明显优于原先的ART2网络.

关 键 词:自适应共振  神经网络  聚类  双向检测  门限
文章编号:1008-973X(2004)12-1540-05
修稿时间:2003年8月7日

ART2 neural network with bidirectional matching mechanism
ZHONG Xu,CHEN De-zhao,CHEN Ya-qiu,LUO Jian-hong.ART2 neural network with bidirectional matching mechanism[J].Journal of Zhejiang University(Engineering Science),2004,38(12):1540-1544.
Authors:ZHONG Xu  CHEN De-zhao  CHEN Ya-qiu  LUO Jian-hong
Abstract:A novel bidirectional matching adaptive resonance theory (BMART2)neural network was proposed to improve insensitivity to gradual processes of traditional ART2(TART2)and original-pattern ART2(OPART2) neural network. By means of adding two bounds, the proposed network had a bidirectional matching on the clustering of input patterns and used output of comparing layer to calculate the degree of similarity. The clustering result of a series of isoceles triangle patterns shows that BMART2 network is sensitive to gradual processes and can perform clustering gradual patterns better than the other two ART2 networks.
Keywords:adaptive resonance  neural network  clustering  bidirectional matching  bound
本文献已被 CNKI 维普 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号