Scholarlink

<p>Multi-view clustering has gained great attention because of its ability to exploit the information obtained from different views, which reflect different properties of data. However, most existing methods only gained hard or fuzzy partitions and cannot accurately represent the uncertainty and imprecision between different views. To address this challenge, we propose a novel Multi-View Deep Evidential C-means (MVDEC) clustering algorithm, which extends the deep evidential clustering framework to multi-view scenarios. MVDEC leverages the strengths of representation learning and belief function theory while integrating complementary information across multiple views. The overall loss function combines view reconstruction loss, a unified evidential clustering loss and regularization term. The evidential partition mechanism allows each object to belong to a single class or any subset of classes, thereby enhancing its ability to model uncertainty. Extensive experiments on real-world multi-view datasets demonstrate that MVDEC outperforms existing multi-view clustering methods and single-view evidential clustering algorithms in terms of clustering accuracy and robustness.</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

A novel deep multi-view clustering algorithm based on feature learning and evidence theory

  • 期刊论文

  • 发表日期：

    2026年4月30日

查看全文
  • Xuan Yang,
  • Fuyuan Xiao

摘要

Multi-view clustering has gained great attention because of its ability to exploit the information obtained from different views, which reflect different properties of data. However, most existing methods only gained hard or fuzzy partitions and cannot accurately represent the uncertainty and imprecision between different views. To address this challenge, we propose a novel Multi-View Deep Evidential C-means (MVDEC) clustering algorithm, which extends the deep evidential clustering framework to multi-view scenarios. MVDEC leverages the strengths of representation learning and belief function theory while integrating complementary information across multiple views. The overall loss function combines view reconstruction loss, a unified evidential clustering loss and regularization term. The evidential partition mechanism allows each object to belong to a single class or any subset of classes, thereby enhancing its ability to model uncertainty. Extensive experiments on real-world multi-view datasets demonstrate that MVDEC outperforms existing multi-view clustering methods and single-view evidential clustering algorithms in terms of clustering accuracy and robustness.