Scholarlink

<p>In recent years, incomplete multi-view partial multi-label classification has attracted growing attention due to its practical relevance. However, many existing methods rely on equal-weight (average) fusion and thus overlook sample-wise reliability differences across views, making the fused representations vulnerable to noise and missing views. To address this issue, we propose the Structure-Aware Consistency Fusion Network (SACF-Net) for incomplete multi-view partial multi-label classification. SACF-Net assigns instance-level view weights based on structural consistency, following the principle that larger structural deviation implies lower view reliability, which enables fine-grained fusion and improves the structural stability of the fused embeddings. Moreover, we convert the prior graph structure into a prior neighborhood distribution and introduce a graph-guided distribution consistency objective to align the neighborhood distribution induced by the fused representation with the prior one, while encouraging agreement among view-specific neighborhood distributions. This design strengthens multi-view consistency at the neighborhood-distribution level and alleviates missingness-induced distribution shifts. Experiments on five public benchmark datasets demonstrate that SACF-Net consistently outperforms competitive methods on multiple evaluation metrics.</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

Incomplete multi–view partial multi–label learning network with structure–aware consistent fusion

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月24日

查看全文
  • Xingang Mao,
  • Yang Xu,
  • Wenbin Bu,
  • Zherui Yang

摘要

In recent years, incomplete multi-view partial multi-label classification has attracted growing attention due to its practical relevance. However, many existing methods rely on equal-weight (average) fusion and thus overlook sample-wise reliability differences across views, making the fused representations vulnerable to noise and missing views. To address this issue, we propose the Structure-Aware Consistency Fusion Network (SACF-Net) for incomplete multi-view partial multi-label classification. SACF-Net assigns instance-level view weights based on structural consistency, following the principle that larger structural deviation implies lower view reliability, which enables fine-grained fusion and improves the structural stability of the fused embeddings. Moreover, we convert the prior graph structure into a prior neighborhood distribution and introduce a graph-guided distribution consistency objective to align the neighborhood distribution induced by the fused representation with the prior one, while encouraging agreement among view-specific neighborhood distributions. This design strengthens multi-view consistency at the neighborhood-distribution level and alleviates missingness-induced distribution shifts. Experiments on five public benchmark datasets demonstrate that SACF-Net consistently outperforms competitive methods on multiple evaluation metrics.