Scholarlink

<p>Sparse Subspace Clustering (SSC) is an efficient model for clustering high-dimensional data. However, most extensions of SSC first learn a similarity matrix and then obtain the clustering results through spectral clustering. These methods require two steps and may suffer from poor performance due to not being globally optimal. Therefore, we propose a semi-supervised sparse subspace clustering method based on label propagation (S4CLP) in this paper. Concretely, we integrate self-representation learning and clustering learning into a joint framework, where the two components interact to improve clustering performance. Moreover, we also incorporate pointwise label constraints and pairwise constraints (cannot-link) into a unified framework. Due to the label information and <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(l_1\)</EquationSource> </InlineEquation>-norm, the coefficient matrix in our model is sparse, which helps alleviate the negative effects of noise and outliers. Additionally, we fully leverage label information to guide the learning of the coefficient matrix and clustering results in our model. This results in a coefficient matrix with strong discriminative power and improved clustering performance. Finally, an efficient algorithm for the proposed model is designed using the Alternating Direction Method of Multipliers (ADMM). Extensive experimental results demonstrate the superiority of S4CLP compared to state-of-the-art clustering models.</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

Semi-supervised sparse subspace clustering based on label propagation

  • 期刊论文

  • 发表日期：

    2026年3月16日

查看全文
  • Miao Fan,
  • Xiafei Yang,
  • Zhiwei Xing

摘要

Sparse Subspace Clustering (SSC) is an efficient model for clustering high-dimensional data. However, most extensions of SSC first learn a similarity matrix and then obtain the clustering results through spectral clustering. These methods require two steps and may suffer from poor performance due to not being globally optimal. Therefore, we propose a semi-supervised sparse subspace clustering method based on label propagation (S4CLP) in this paper. Concretely, we integrate self-representation learning and clustering learning into a joint framework, where the two components interact to improve clustering performance. Moreover, we also incorporate pointwise label constraints and pairwise constraints (cannot-link) into a unified framework. Due to the label information and \(l_1\) -norm, the coefficient matrix in our model is sparse, which helps alleviate the negative effects of noise and outliers. Additionally, we fully leverage label information to guide the learning of the coefficient matrix and clustering results in our model. This results in a coefficient matrix with strong discriminative power and improved clustering performance. Finally, an efficient algorithm for the proposed model is designed using the Alternating Direction Method of Multipliers (ADMM). Extensive experimental results demonstrate the superiority of S4CLP compared to state-of-the-art clustering models.