基于集群的并行分布式聚类及其应用(英文)  

Cluster-PC Based Parallel Distributed Data Clustering and Its Applications

作　　者：夏胜平[1] 吕小军[1] 刘建军[1] 袁振涛[1] 郁文贤[1] 

机构地区：[1]国防科学技术大学电子科学与工程学院ATR重点实验室,长沙410073

出　　处：《郑州大学学报（理学版）》2006年第4期33-40,共8页Journal of Zhengzhou University:Natural Science Edition

基　　金：武器装备预研基金资助项目

摘　　要：海量和高维大数据集的聚类对计算机性能提出了很高的要求.基于具有层次聚类特性的RSOM树方法提供了一种有效的手段以实现对高维大数据集的聚类索引,这种RSOM树可支持最近邻搜索且不需要对数据进行线性搜索.注意到RSOM模型具有内在的层次化、分布式结构特点,并可进行增量的训练,研究了基于高效并行集群的增量、分布式RSOM并行算法,并通过视频图像特征集实例证实了算法的可行性.Clustering data with high dimensionalities requires high-performance computers to get results in a reasonable amount of time, particularly for extremely large-scale databases. Thus, the recursive SOM（RSOM） tree method is proposed. RSOM tree is a hierarchy of clusters and sub clusters which incorporates the cluster representation into the index structure. It provides a practical solution to index clustered data set, and it supports the retrieval of the nearest-neighbors effectively and efficiently without having to linearly search a high-dimensional large database. Meanwhile, an incremental RSOM tree-based clustering algorithm is proposed; and because of the RSOM tree is of the nature of parallelism, and can be implemented on scalable parallel computers. Thus a cluster-system based distributed parallel algorithm of incremental RSOM tree is proposed. The performance of the method has been tested with high dimensional feature sets extracted from large image database.

关 键 词：并行分布式聚类 RSOM 集群系统 增量聚类 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]