• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 论文 • 上一篇    下一篇

一种利用极限学习机的数据可视化方法

陈文兵,宋玛君,王廷春   

  1. (南京信息工程大学数学与统计学院,江苏 南京 210044)
  • 收稿日期:2015-12-11 修回日期:2016-02-23 出版日期:2017-05-25 发布日期:2017-05-25
  • 基金资助:

    国家自然科学基金(1157118);国家行业专项基金(201406029,201306043);北极阁基金(BJG201504)

A data visualization method
based on extreme learning machine

CHEN Wen-bing,SONG Ma-jun,WANG Ting-chun   

  1. (School of Mathematics and Statistics,Nanjing University of Information Science and Technology,Nanjing 210044,China)
  • Received:2015-12-11 Revised:2016-02-23 Online:2017-05-25 Published:2017-05-25

摘要:

提出一种利用极限学习机ELM的数据可视化方法,该方法利用多维尺度分析MDS、Pearson相关性、Spearman相关性代替常用的均方误差MSE实现高维数据投影到2-维平面的数据可视化。将所提方法与近期流行的随机邻域嵌入SNE及其改进的t-SNE方法对比,并通过局部连续元准则LCMC进行质量评测。结果表明:该方法的数据可视化结果及计算性能明显优于SNE及t-SNE方法;而在提出的三种学习规则中,基于MDS的学习规则效果最好。
 

关键词: 数据可视化, ELM, MDS, Pearson相关性, Spearman相关性

Abstract:

We present a novel data visualization method based on the extreme learning machine (ELM), which uses the multidimensional scaling (MDS), Pearson correlation and Spearman correlation respectively instead of the common MSE to project the high-dimensional data onto a two-dimensional plane to carry out data visualization. Experimental results show that compared with the recent popular stochastic neighbor embedding (SNE) and t-SNE, the proposed method outperforms them in visual effect and computation performance. Furthermore, these experimental results also show that the MDS-based ELM has the best performance.
 

Key words: data visualization, ELM, MDS, Pearson correlation, Spearman correlation