基于多级自表示约束的不完备多视图聚类  

Incomplete multi-view clustering based on multi-level self-representation constraints

作　　者：陈梅[1] 马学艳 钱罗雄 张锦宏 张弛 CHEN Mei;MA Xue-yan;QIAN Luo-xiong;ZHANG Jin-hong;ZHANG Chi(School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730000,China)

机构地区：[1]兰州交通大学电子与信息工程学院,兰州730000

出　　处：《控制与决策》2025年第2期645-654,共10页Control and Decision

基　　金：国家自然科学基金项目(62266029);甘肃省高等学校产业支撑计划项目(2022CYZC-36)。

摘　　要：针对现有的不完备多视图聚类方法存在无法准确利用缺失视图的潜在信息和未能充分利用视图间的互补信息以及高阶相关性等问题,提出一种新的基于多级自表示约束的不完备多视图聚类(CMLC)方法.CMLC利用公共潜在表示来恢复缺失值,从而有效获取缺失部分的潜在信息.为了获得多视图数据的统一低秩表示,CMLC首先通过多级自表示约束捕获多视图数据内部的一致信息和视图间的互补信息,同时利用多级误差表示提高模型对噪声的鲁棒性,接着通过张量对数行列式捕获视图间的高阶相似信息,最后引入距离正则项捕获数据的局部信息.与9个对比方法在多种缺失率下的6个仿真不完备多视图数据集上进行实验对比,结果表明CMLC均获得了最好的聚类性能.The existing incomplete multi-view clustering methods have some problems,such as failing to make full use of the potential information of missing views,and failing to make full use of the complementary information and high order correlation among views.In this paper,a new incomplete multi-view clustering method based on multi-level self representation constraints(CMLC)is proposed.The CMLC uses the common latent representation to recover the missing value and thus effectively obtain the latent information of the missing part.In order to obtain a unified low-rank representation of multi-view data,the CMLC first captures the consistent information within the multi-view data and the complementary information between views through multi-level self-representation constraints.At the same time,it uses multi-level error representation to improve the robustness of the model against noise,and then captures the higher-order similar information between views through the logarithmic tensor.Finally,the distance regular term is introduced to capture the local information of the data.The results show that the CMLC has the best clustering performance compared with nine methods on six imperfectly simulated multi-view datasets with different miss rates.

关 键 词：不完备多视图聚类 一致表示 张量对数行列式 低秩张量 不完备数据 张量分析 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]