一种改进的高维数据可视化模型  

A Modified Visualization-Model of High-dimensional Data

作　　者：彭红毅[1] 蒋春福[2] 朱思铭[3] 

机构地区：[1]华南农业大学理学院,广州510642 [2]深圳大学数学与计算科学学院,深圳518060 [3]中山大学数学与计算科学学院,广州510275

出　　处：《计算机科学》2007年第4期175-178,共4页Computer Science

基　　金：国家自然科学基金资助(10371135)

摘　　要：可视化诱导自组织映射(ViSOM)是一种人工神经网络模型,已经被成功应用于高维数据的可视化分析。但是,标准的ViSOM方法不仅没有考虑数据之间的相关性,而且当输出网络结点太多时,需要消耗大量运算开销;输出网络结点太少,又难以分析数据的可视化结果。为克服ViSOM的这两个弱点,本文首先在ViSOM的基础上提出了一个改进的映射算法MViSOM,接着在独立成分分析(ICA)与MViSOM的基础上提出了一个改进的高维数据可视化模型IMViSOM。论文最后通过实验说明了IMViSOM模型在对群聚数据的可视化分类效果及运算速度方面都优于ViSOM方法,从而验证了IMViSOM模型的正确性与合理性。The Visualization-Induced Self-Organizing Maps （ViSOM）, as one of the artificial neural networks models, has been successfully applied in the analysis of visualization of high-dimensional data. However, it has two weaknesses. Firstly, it does not consider the correlation of data. Secondly, much memory will be used up if the output nodes are too large, and contrarily, the visibility results of data will be difficult to be analyzed if the output nodes are too small. In order to overcome the above two weaknesses of ViSOM, a modified algorithm named MViSOM, based on ViSOM, as well as a visualization-model of high-dimensional data, based on ICA （Independent Component Analysis）and MViSOM, are proposed in this paper. Finally, the experiments also show that IMViSOM method has advantages over ViSOM because of its excellent classified effect of swarm data and high calculating speed, confirming the correctness and reasonableness for the proposed model in this paper.

关 键 词：独立成分分析 可视化诱导自组织映射 相关性 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]