《计算机应用》唯一官方网站 ›› 2021, Vol. 41 ›› Issue (12): 3468-3474.DOI: 10.11772/j.issn.1001-9081.2021061393

• 第十八届中国机器学习会议(CCML 2021) • 上一篇    下一篇

基于上界单纯形投影图张量学习的多核聚类算法

雷皓云1,2, 任珍文1,3(), 汪彦龙4, 薛爽1, 李浩然1   

  1. 1.西南科技大学 国防科技学院,四川 绵阳 621010
    2.电子科技大学 信息与通信工程学院,成都 611731
    3.计算机软件新技术国家重点实验室(南京大学),南京 210023
    4.浙江传媒学院 媒体工程学院,浙江 杭州 310018
  • 收稿日期:2021-05-12 修回日期:2021-08-30 接受日期:2021-08-31 发布日期:2021-12-28 出版日期:2021-12-10
  • 通讯作者: 任珍文
  • 作者简介:雷皓云(1999—),男,四川成都人,硕士研究生,CCF会员,主要研究方向:多核聚类、雷达成像技术
    汪彦龙(1971—),男,黑龙江黑河人,硕士,主要研究方向:医学图像处理
    薛爽(2000—),女,山西运城人,主要研究方向:多核聚类、多视图聚类
    李浩然(1997—),男,河北石家庄人,硕士研究生,主要研究方向:多核聚类、多视图聚类。
  • 基金资助:
    四川省科技厅应用基础研究项目(2021YJ0083);国家自然科学基金资助项目(62106209);南京大学计算机软件新技术国家重点实验室资助项目(KFKT2021B23);浙江省基础公益研究计划项目(LGF21F020003);重庆自然科学基金资助项目(cstc2020jcyj-msxmX0473);浙江省影视媒体技术研究重点实验室开放基金课题(2020E10015)

Multiple kernel clustering algorithm based on capped simplex projection graph tensor learning

Haoyun LEI1,2, Zhenwen REN1,3(), Yanlong WANG4, Shuang XUE1, Haoran LI1   

  1. 1.School of National Defense Science and Technology,Southwest University of Science and Technology,Mianyang Sichuan 621010,China
    2.School of Information and Communication Engineering,University of Electronic Science and Technology of China,Chengdu Sichuan 611731,China
    3.State Key Laboratory for Novel Software Technology (Nanjing University),Nanjing Jiangsu 210023,China
    4.School of Media Engineering,Communication University of Zhejiang,Hangzhou Zhejiang 310018,China
  • Received:2021-05-12 Revised:2021-08-30 Accepted:2021-08-31 Online:2021-12-28 Published:2021-12-10
  • Contact: Zhenwen REN
  • About author:LEI Haoyun, born in 1999, M. S. candidate. His research interests include multiple kernel clustering, radar imaging technology.
    WANG Yanlong, born in 1971. M. S. His research interests include medical image processing.
    XUE Shuang, born in 2000. Her research interests include multiple kernel clustering, multi-view clustering.
    LI Haoran, born in 1997, M. S. candidate. His research interests include multiple kernel clustering, multi-view clustering.
  • Supported by:
    the Application Foundation Research Project of Science and Technology Department of Sichuan Province(2021YJ0083);the National Natural Science Foundation of China(62106209);the State Key Lab for Novel Software Technology Project at Nanjing University(KFKT2021B23);the Basic Public Welfare Research Program of Zhejiang Province(LGF21F020003);the Natural Science Foundation of Chongqing(cstc2020jcyj-msxmX0473);the Open Fund Project of Key Lab of Film and TV Media Technology of Zhejiang Province(2020E10015)

摘要:

近年来,多核图聚类(MKGC)受到了广泛的关注,这得益于多核学习能有效地避免核函数与核参数的选择,而图聚类能充分挖掘样本间的复杂结构信息。然而现有的MKGC方法存在着如下问题:图学习技术使得模型复杂化,图拉普拉斯矩阵的高秩特性使其难以保证学到的关系图包含精确的c个连通分量(块对角性质),以及大部分方法忽略了候选关系图间的高阶结构信息,使得多核信息难以被充分利用。针对以上问题,提出了一种新的MKGC方法。首先,提出一种新的上界单纯形投影图学习方法,直接将核矩阵投影到图单纯形上,降低了计算复杂度;同时,引入一种新的块对角约束,使学到的关系图能保持精确的块对角属性;此外,在上界单纯形投影空间中引入低秩张量学习来充分挖掘多个候选关系图的高阶结构信息。在多个数据集上与现有的MKGC方法相比,所提出方法计算量小、稳定性高,在聚类精度(ACC)和标准互信息(NMI)指标上具有较大的优势。

关键词: 多核图聚类, 上界单纯形, 张量学习, 块对角性质, 高阶结构信息

Abstract:

Because multiple kernel learning can avoid selection of kernel functions and parameters effectively, and graph clustering can fully mine complex structural information between samples, Multiple Kernel Graph Clustering (MKGC) has received widespread attention in recent years. However, the existing MKGC methods suffer from the following problems: graph learning technique complicates the model, the high rank of graph Laplacian matrix cannot ensure the learned affinity graph to contain accurate c connected components (block diagonal property), and most of the methods ignore the high-order structural information among the candidate affinity graphs, making it difficult to fully utilize the multiple kernel information. To tackle these problems, a novel MKGC method was proposed. First, a new graph learning method based on capped simplex projection was proposed to directly project the kernel matrices onto graph simplex, which reduced the computational complexity. Meanwhile, a new block diagonal constraint was introduced to keep the accurate block diagonal property of the learned affinity graphs. Moreover, the low-rank tensor learning was introduced in capped simplex projection space to fully mine the high-order structural information of multiple candidate affinity graphs. Compared with the existing MKGC methods on multiple datasets, the proposed method has less computational cost and high stability, and has great advantages in Accuracy (ACC) and Normalized Mutual Information (NMI).

Key words: Multiple Kernel Graph Clustering (MKGC), capped simplex, tensor learning, block diagonal property, high-order structural information

中图分类号: