一种生长型自组织神经网络的聚类研究  

Clustering Study of a Growing Self-Organizing Neural Network

作　　者：傅雪[1] 张少白[1] 

机构地区：[1]南京邮电大学计算机学院,江苏南京210003

出　　处：《计算机技术与发展》2011年第3期64-66,71,共4页Computer Technology and Development

基　　金：山东省自然科学基金(Y2007G34;Y2006G03);南京邮电大学引进人才基金项目(NY207134)

摘　　要：自组织特征映射神经网络SOM(Self-Organizing Feature Maps)是一种优良的聚类工具,但其存在着一些限制,如需要预先定义网络大小、网络的收敛性较差和结构不灵活等。为了克服这些不足,在自组织神经网络理论的指导下,提出了一种基于生长型自组织神经网络的聚类方法。在无监督的情况下,该方法采用阈值控制的触发机制实现网络中神经元的生长和删除,并通过神经元权值的有效调整,以期得到数据对象的聚类结果。实验以二维空间中的数据对象为输入样本,验证了该方法的有效性和优越性。The self-organizing feature maps is a good clustering tool, but there are some restrictions, such as it needs to pre-define the network size, its convergence is poor mad the structure is not flexible. To overcome these shortcomings, a clustering method based on a growing self-organizing neural network is proposed by the knowledge of self-organizing neural network. This method controls neural＇s growths and deletions by implementing trigger mechanism of the threshold value without supervision, and through making adjustments of neural weight, it can get clustering results of data objects. The experiment results prove the method＇s effectiveness and superiority by choosing data objects in two-dimensional space as input samples.

关 键 词：自组织 生长 特征映射 聚类 神经网络 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]