Scholarlink

<p>Unsupervised feature extraction (UFE) has attracted increasing attention in machine learning, data mining and pattern recognition, as it effectively uncovers the intrinsic low-dimensional structure of high-dimensional data. However, most existing UFE methods focus on either global structures or local structures, with few algorithms successfully balancing the two. In real-world applications, data frequently exhibit complex structures and noise, so learning models that rely solely on a single structure of data can easily lead to overfitting or introducing unnecessary biases, which greatly limits their applications. To address the issues, we propose a novel UFE method, called Adaptive Structure Graph Embedding (ASGE). ASGE jointly captures global and local structures by integrating distance constraints into low-rank representation learning. Specifically, it employs an&#xa0;<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10489_2025_6746_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(\ell _2\)</EquationSource> </InlineEquation> Frobenius norm regularizer to model global correlations and an <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10489_2025_6746_Article_IEq2.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(\ell _1\)</EquationSource> </InlineEquation> norm regularizer to enhance robustness against noise and outliers. In addition, ASGE imposes the nonnegative constraint on the representation to make the learned graph interpretable and reveal the intrinsic structure of the data. The experimental results on real-world datasets show that ASGE outperforms other 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

Adaptive structure graph embedding for unsupervised feature extraction

  • 期刊论文

  • 发表日期：

    2025年7月10日

查看全文
  • Qiqi Chen,
  • Xuesong Yin,
  • Jianhao Ding,
  • Qi Huang,
  • Ting Shu,
  • Yigang Wang

摘要

Unsupervised feature extraction (UFE) has attracted increasing attention in machine learning, data mining and pattern recognition, as it effectively uncovers the intrinsic low-dimensional structure of high-dimensional data. However, most existing UFE methods focus on either global structures or local structures, with few algorithms successfully balancing the two. In real-world applications, data frequently exhibit complex structures and noise, so learning models that rely solely on a single structure of data can easily lead to overfitting or introducing unnecessary biases, which greatly limits their applications. To address the issues, we propose a novel UFE method, called Adaptive Structure Graph Embedding (ASGE). ASGE jointly captures global and local structures by integrating distance constraints into low-rank representation learning. Specifically, it employs an  \(\ell _2\) Frobenius norm regularizer to model global correlations and an \(\ell _1\) norm regularizer to enhance robustness against noise and outliers. In addition, ASGE imposes the nonnegative constraint on the representation to make the learned graph interpretable and reveal the intrinsic structure of the data. The experimental results on real-world datasets show that ASGE outperforms other state-of-the-art methods.