Scholarlink

<p>Multi-label learning aims to exploit label correlation for prediction. In this paper, we learn operator-valued kernels from multi-label dataset to achieve finer-grained characterization of label correlation. Firstly, the global importance distribution of feature set to label is calculated based on Hilbert–Schmidt Independence Criterion (HSIC). The construction of separable operator-valued kernel relies on these importance distributions to describe the global label correlation. Secondly, the instance-level feature importance distribution is further learned to develop transformable operator-valued kernel by using HSIC. The block operator kernel matrix of transformable kernel describes the instance-level label correlation. Thirdly, multi-label learning algorithms associated with operator-valued kernel are designed to tackle multi-label learning prediction tasks. In order to demonstrate the effectiveness of our proposed algorithms, the classification experiments and statistical analysis results on eight multi-label datasets are presented, and the results are compared with five high-performance algorithms.</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

Multi-label learning based on operator-valued kernels

  • 期刊论文

  • 发表日期：

    2025年7月10日

查看全文
  • Zhenxin Wang,
  • Degang Chen,
  • Xiaoya Che

摘要

Multi-label learning aims to exploit label correlation for prediction. In this paper, we learn operator-valued kernels from multi-label dataset to achieve finer-grained characterization of label correlation. Firstly, the global importance distribution of feature set to label is calculated based on Hilbert–Schmidt Independence Criterion (HSIC). The construction of separable operator-valued kernel relies on these importance distributions to describe the global label correlation. Secondly, the instance-level feature importance distribution is further learned to develop transformable operator-valued kernel by using HSIC. The block operator kernel matrix of transformable kernel describes the instance-level label correlation. Thirdly, multi-label learning algorithms associated with operator-valued kernel are designed to tackle multi-label learning prediction tasks. In order to demonstrate the effectiveness of our proposed algorithms, the classification experiments and statistical analysis results on eight multi-label datasets are presented, and the results are compared with five high-performance algorithms.