Scholarlink

<p>Subspace clustering has garnered significant attention for its commendable interpretability and performance. Existing methods primarily concentrate on two vital components: the effective similarity matrix construction and the sparse projection matrix optimization. However, these methods encounter some challenges. Firstly, they struggle with large-scale datasets because constructing the similarity matrix based on all samples incurs prohibitive computational costs during the optimization process. Secondly, they usually involve two separate stages, i.e., learning the low-dimensional representation and then discretizing it, which may bias the optimal solutions. In this study, we propose a self-weighted anchor-based scalable subspace clustering (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({\text {WA}}{{\text {S}}^{\text {2}}}{\text {C}}\)</EquationSource> </InlineEquation>) model, in which an anchor-based strategy is used to improve model scalability for large-scale data sets. More importantly, to improve the representative ability of the selected anchor points, we introduce a self-weighted anchor mechanism to quantitatively measure the contribution of each anchor and then guarantee the quality of the corresponding anchor graph. Furthermore, we integrate the optimization of low-dimensional representations and discrete indicator matrices into a unified framework, thereby mitigating the error accumulation inherent in two-stage approaches. Comprehensive experimentation on benchmark datasets validate the superior performance of the proposed model in contrast to several state-of-the-art methods.</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

Self-weighted anchor-based scalable subspace clustering

  • 期刊论文

  • 发表日期：

    2026年5月16日

查看全文
  • Chucheng Huang,
  • Jie Zhou,
  • Yue Guo,
  • Can Gao,
  • Witold Pedrycz

摘要

Subspace clustering has garnered significant attention for its commendable interpretability and performance. Existing methods primarily concentrate on two vital components: the effective similarity matrix construction and the sparse projection matrix optimization. However, these methods encounter some challenges. Firstly, they struggle with large-scale datasets because constructing the similarity matrix based on all samples incurs prohibitive computational costs during the optimization process. Secondly, they usually involve two separate stages, i.e., learning the low-dimensional representation and then discretizing it, which may bias the optimal solutions. In this study, we propose a self-weighted anchor-based scalable subspace clustering ( \({\text {WA}}{{\text {S}}^{\text {2}}}{\text {C}}\) ) model, in which an anchor-based strategy is used to improve model scalability for large-scale data sets. More importantly, to improve the representative ability of the selected anchor points, we introduce a self-weighted anchor mechanism to quantitatively measure the contribution of each anchor and then guarantee the quality of the corresponding anchor graph. Furthermore, we integrate the optimization of low-dimensional representations and discrete indicator matrices into a unified framework, thereby mitigating the error accumulation inherent in two-stage approaches. Comprehensive experimentation on benchmark datasets validate the superior performance of the proposed model in contrast to several state-of-the-art methods.