面向深度不完整多视图聚类的失衡视图贡献评估与精炼
Imbalanced View Contribution Evaluation and Refinement for Deep Incomplete Multi-View Clustering
用 Shapley 值量化各视图的边际贡献、并用非平衡最优传输刻画视图间分布偏移，这套「贡献评估+协作增强」的思路可迁移到多模态/多视图特征融合与缺失模态鲁棒表征学习中，帮助识别并抑制主导模态对融合的压制。
原文依据 we employ Shapley values to quantify each view's marginal contribution, and incorporate unbalanced optimal transport to characterize distributional deviations across views
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论在实际应用中，多视图数据常因隐私保护和传感器故障而出现缺失情况。这类不完整场景不仅降低了信息可用性，还导致视图之间出现显著失衡：某些「强视图」主导融合过程，而「弱视图」贡献甚微，从而削弱了跨视图协作。现有的不完整多视图聚类方法主要关注缺失数据的处理，却忽视了不完整性所引发的视图贡献失衡及其对表征学习和聚类性能的影响。为解决该问题，本文首先分析了缺失视图造成的数据失衡以及由此产生的视图学习质量差异。随后，我们提出了一种面向失衡不完整多视图聚类的失衡贡献评估与精炼框架（ICER）。具体而言，我们采用 Shapley 值量化每个视图的边际贡献，并引入非平衡最优传输来刻画视图间的分布偏移。在此基础上，我们构建了视图贡献失衡指标，以全面评估跨视图协作与融合质量，并设计了一个协作增强模块，以显式强化视图间的协同优化。在多个数据集上的大量实验表明，所提方法优于现有的不完整多视图聚类方法，验证了在失衡不完整场景中显式建模并缓解视图失衡的有效性与必要性。代码已发布于 https://github.com/Evelyn-zhou24/ICER。
英文原文摘要 1652 字符
In real-world applications, multi-view data often suffer from missing situations due to privacy protection and sensor failures. Such incomplete scenarios not only reduce information availability but also cause significant imbalance among views: certain "strong views" dominate the fusion process, while "weak views" contribute marginally, undermining cross-view collaboration. Existing incomplete multi-view clustering methods mainly focus on handling missing data, but overlook the view contribution imbalance induced by incompleteness and its impact on representation learning and clustering performance. To address this issue, we first analyze the data imbalance caused by missing views and the resulting disparities in view learning quality. Then, we propose an Imbalanced Contribution Evaluation and Refinement framework (ICER) for imbalanced incomplete multi-view clustering. Specifically, we employ Shapley values to quantify each view's marginal contribution, and incorporate unbalanced optimal transport to characterize distributional deviations across views. On this basis, we construct the view contribution imbalance metric to comprehensively evaluate cross-view collaboration and fusion quality, and design a collaboration enhancement module to explicitly reinforce inter-view cooperative optimization. Extensive experiments on multiple datasets demonstrate that the proposed method outperforms existing incomplete multi-view clustering approaches, validating the effectiveness and necessity of explicitly modeling and mitigating view imbalance in imbalanced incomplete scenarios. Code is released at https://github.com/Evelyn-zhou24/ICER.