基于集群的并行分布式聚类及其应用_英文_.pdfVIP

　第 38 卷第 4 期 ( ) Vol38 No4 郑 州 大 学 学 报 理 学 版 　 2006 年 12 月 J . of Zhengzho u U niv . (N at . Sci . Ed . ) Dec 2006 Cl usterPC Based Parallel Distributed Data Cl ustering and Its Appl ications XIA Shengp ing , 　L U¨Xiaoj un , 　L IU J ianj un , YU AN Zhent ao , 　YU Wenxian (A TR S t ate Key L aboratory , N at ional Uni v ers i ty of D ef ens e Tech nol og y , Ch angs h a 4 10073 , Ch i na) Abstract :Clu st ering dat a wit h high dimen sionalities requires highp erformance comp ut er s to get result s in a rea sonable amount of time , p articularly for ext remely lar gescale dat aba ses. Thu s , t he recur sive SOM ( R SOM) t ree met hod i s p ropo sed . R SOM t ree i s a hierarchy of clu st er s and subclu st er s which incorporat es t he clu st er rep resent ation into t he index st ruct ure . It p rovides a p ractical solution to index clu st ered dat a set , and it support s t he ret rieval of t he nearestneighbor s effectively and efficiently wit hout having to linearly search a highdimen sional lar ge dat abase . Meanwhile , an increment al R SOM t reebased clu st ering algorit hm i s p ropo sed ; and becau se of t he R SOM t ree i s of t he nat ure of p aralleli sm , and can be implement ed on scalable p arallel comp ut er s. Thu s a clu st ersyst em based di st ribut ed p arallel algorit hm of increment al R SOM t ree i s p ropo sed . The p erformance of t he met hod has been t est ed wit