Efficient Multiple Kernel Clustering via Spectral Perturbation

被引:6
作者
Tang, Chang [1 ,2 ]
Li, Zhenglai [1 ,2 ]
Yan, Weiqing [3 ]
Yue, Guanghui [4 ]
Zhang, Wei [5 ]
机构
[1] China Univ Geosci, Sch Comp Sci, Wuhan 430074, Peoples R China
[2] Nanjing Univ, State Key Lab Novel Software Technol, Nanjing 210023, Peoples R China
[3] Yantai Univ, Sch Comp & Control Engn, Yantai 264005, Peoples R China
[4] Shenzhen Univ, Hlth Sci Ctr, Sch Biomed Engn, Shenzhen 518060, Peoples R China
[5] Qilu Univ Technol, Shandong Acad Sci, Shandong Prov Key Lab Comp Networks, Shandong Comp Sci Ctr,Natl Supercomp Ctr Jinan, Jinan 250353, Peoples R China
来源
PROCEEDINGS OF THE 30TH ACM INTERNATIONAL CONFERENCE ON MULTIMEDIA, MM 2022 | 2022年
基金
美国国家科学基金会;
关键词
Multiple kernel clustering; spectral perturbation; partition fusion; GRAPH;
D O I
10.1145/3503161.3548153
中图分类号
TP39 [计算机的应用];
学科分类号
081203 ; 0835 ;
摘要
Clustering is a fundamental task in the machine learning and data mining community. Among existing clustering methods, multiple kernel clustering (MKC) has been widely investigated due to its effectiveness to capture non-linear relationships among samples. However, most of the existing MKC methods bear intensive computational complexity in learning an optimal kernel and seeking the final clustering partition. In this paper, based on the spectral perturbation theory, we propose an efficient MKC method that reduces the computational complexity from O(n(3)) to O(nk(2) + k(3)), with n and k denoting the number of data samples and the number of clusters, respectively. The proposed method recovers the optimal clustering partition from base partitions by maximizing the eigen gaps to approximate the perturbation errors. An equivalent optimization objective function is introduced to obtain base partitions. Furthermore, a kernel weighting scheme is embedded to capture the diversity among multiple kernels. Finally, the optimal partition, base partitions, and kernel weights are jointly learned in a unified framework. An efficient alternate iterative optimization algorithm is designed to solve the resultant optimization problem. Experimental results on various benchmark datasets demonstrate the superiority of the proposed method when compared to other state-of-the-art ones in terms of both clustering efficacy and efficiency.
引用
收藏
页码:1603 / 1611
页数:9
相关论文
共 39 条
  • [1] [Anonymous], 2014, ADV NEURAL INFORM PR
  • [2] Cortes C., 2013, ICML, P46, DOI DOI 10.1007/978-1-4471-4351-2_3
  • [3] Cortes C, 2012, J MACH LEARN RES, V13, P795
  • [4] Du L, 2015, PROCEEDINGS OF THE TWENTY-FOURTH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE (IJCAI), P3476
  • [5] Consensus graph and spectral representation for one-step multi-view kernel based clustering
    El Hajjar, S.
    Dornaika, F.
    Abdallah, F.
    Barrena, N.
    [J]. KNOWLEDGE-BASED SYSTEMS, 2022, 241
  • [6] Gao LL, 2016, AAAI CONF ARTIF INTE, P1188
  • [7] Multiple Kernel Fuzzy Clustering
    Huang, Hsin-Chien
    Chuang, Yung-Yu
    Chen, Chu-Song
    [J]. IEEE TRANSACTIONS ON FUZZY SYSTEMS, 2012, 20 (01) : 120 - 134
  • [8] Huang L, 2013, 2013 1ST INTERNATIONAL FUTURE ENERGY ELECTRONICS CONFERENCE (IFEEC 2013), P431, DOI 10.1109/IFEEC.2013.6687544
  • [9] Hunter Blake, 2010, ARXIV10110997
  • [10] Structured graph learning for clustering and semi-supervised classification
    Kang, Zhao
    Peng, Chong
    Cheng, Qiang
    Liu, Xinwang
    Peng, Xi
    Xu, Zenglin
    Tian, Ling
    [J]. PATTERN RECOGNITION, 2021, 110