Scholarlink

<p>To extract more comprehensive and complete multi-view data representation information from different perspectives and strengthen the learning power of multi-view learning models, a novel enhanced multi-view subspace clustering via dual-norm and low-rank tensor constraints (EMSCDLT) is developed in this study. To uncover the high-order relationships hidden within multi-view data, this model stacks the subspace representation matrix of each view into a tensor and removes redundant information in the data representation using the low-rank (LR) constraint of the tensor. Simultaneously, a Frobenius norm constraint is introduced to strengthen the intra-class data association, and a novel sparse constraint is imposed on the transposition of the subspace representation matrix and its own product to strengthen its block diagonal structure, improving the recognition ability of the model. For the EMSCDLT model, we developed an efficient solution algorithm and analyzed its convergence. Experiments were conducted on multiple challenging datasets, and the results verified the superiority of the proposed method.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced multi-view subspace clustering via dual-norm and low-rank tensor constraints

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Guoqing Liu,
  • Xiujian Hu,
  • Piao Shi,
  • Fujin Chen,
  • Yueguang Li

摘要

To extract more comprehensive and complete multi-view data representation information from different perspectives and strengthen the learning power of multi-view learning models, a novel enhanced multi-view subspace clustering via dual-norm and low-rank tensor constraints (EMSCDLT) is developed in this study. To uncover the high-order relationships hidden within multi-view data, this model stacks the subspace representation matrix of each view into a tensor and removes redundant information in the data representation using the low-rank (LR) constraint of the tensor. Simultaneously, a Frobenius norm constraint is introduced to strengthen the intra-class data association, and a novel sparse constraint is imposed on the transposition of the subspace representation matrix and its own product to strengthen its block diagonal structure, improving the recognition ability of the model. For the EMSCDLT model, we developed an efficient solution algorithm and analyzed its convergence. Experiments were conducted on multiple challenging datasets, and the results verified the superiority of the proposed method.