计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (5): 81-84.DOI: 10.3778/j.issn.1002-8331.1507-0300

• 大数据与云计算 • 上一篇    下一篇

应用谱回归和图正则最小二乘回归的数据降维

翁  谦1,2,3,毛政元2,林嘉雯3,简彩仁4   

  1. 1.福州大学 经济与管理学院,福州 350116
    2.福州大学 福建省空间信息工程研究中心,福州 350002
    3.福州大学 数学与计算机科学学院,福州 350108
    4.厦门大学 嘉庚学院,福建 漳州 363105
  • 出版日期:2017-03-01 发布日期:2017-03-03

Dimension reduction method based on spectral regression and graph regularization least square regression

WENG Qian1,2,3, MAO Zhengyuan2, LIN Jiawen3, JIAN Cairen4   

  1. 1.School of Economics & Management, Fuzhou University, Fuzhou 350116, China
    2.Spatial Information Research Center of Fujian Province, Fuzhou University, Fuzhou 350002, China
    3.College of Mathematics and Computer Science, Fuzhou University, Fuzhou 350108, China
    4.Tan Kah Kee College, Xiamen University, Zhangzhou, Fujian 363105, China
  • Online:2017-03-01 Published:2017-03-03

摘要: 数据降维对于提高高维数据处理的效率具有重要意义,稀疏编码是目前受到广泛关注的主流降维方法。针对该方法在降维过程中不能保持样本空间几何结构信息的不足,提出一种基于谱回归和图正则最小二乘回归的改进方案,以2个图像数据集和2个基因表达数据集为样本的实验表明该方法优于未加改进的稀疏编码降维法。

关键词: 谱回归, 图正则最小二乘回归, 降维, 聚类

Abstract: Data dimension reduction is significant to research high-dimensional data. Sparse concept coding receives widespread attention, but the sparse representation coefficients fail to maintain the essential structure of the data. In response to this discovery, a method based on spectral regression and graph regularization least square regression for data dimension reduction is proposed. The experiments on two image data sets and two gene expression data sets show the proposed method is better than the unimproved sparse concept coding.

Key words: spectral regression, graph regularization least square regression, dimension reduction, clustering