基于统一测量和张量学习的多视图无监督特征选择 米

打开文本图片集

中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2026）04-017-1112-08

doi：10.19734/j. issn.1001-3695.2025.08.0297

Multi-view unsupervised feature selection based on unified measurement and tensor learning

Dai Jiamin，Xie Xijiong† （School of InformationScienceand Engineering，Ningbo University，Ningbo Zhejiang，China）

Abstract：Multi-viewdataareincreasinglycommoninhigh-dimensional industrialapplications.Featureselectionisimportant asitpreservestheoriginalmeaningandinterpretabilityoffeatures.Thisisespeciallusefulwhenlabelsarescarce，making multi-viewunsupervisedfeatureselection（MvUFS）highlypractical.Existing methodsoftenfallshortinexploringinter-view relationshipsandconsistency.Toovercomethisshortage，this paperproposedanewmethodcaledSMUMT.Itintegratedselfrepresentationlearningtoimprovesamplerepresentation.Italsoused jointlearning tobuild areliablesimilaritygraphforguidingfeatureselection.Aditionally，itintroducedtensorlearning tomodelhighordercorrelationsacrossviews.Itconucted clustering experiments onseven publicdatasets.Resultsshow that SMUMToutperforms six state-of-the-art methods inmost cases.Itperformedparticularlywellonimagedatasets.Thesefindingsconfirmthatthismethodisefectiveforfeatureselection and improves clustering performance.

Key words： tensor learning；self-representation learning；unsupervised feature selection

0引言

随着数据采集技术的进步，数据通常使用不同的特征描述符来描述或通过多个信源收集，此类数据称为多视图数据[1]，在各种工业应用中普遍存在，通常具有高维特征[2]。（剩余16601字）