Multi-view clustering (MvC) aims at exploring the category structure among multi-view data without label supervision. Multiple views provide more information than single views and thus existing MvC methods can achieve satisfactory performance. However, their performance might seriously degenerate when the views are noisy in practical scenarios. In this paper, we first formally investigate the drawback of noisy views and then propose a theoretically grounded deep MvC method (namely MvCAN) to address this issue. Specifically, we propose a novel MvC objective that enables un-shared parameters and inconsistent clustering predictions across multiple views to reduce the side effects of noisy views. Furthermore, a non-parametric iterative process is designed to generate a robust learning target for mining multiple views' useful information. Theoretical analysis reveals that MvCAN works by achieving the multi-view consistency, complementarity, and noise robustness. Finally, experiments on public datasets demonstrate that MvCAN outperforms state-of-the-art methods and is robust against the existence of noisy views.

翻译：多视图聚类（MvC）旨在无标签监督下探索多视图数据的类别结构。由于多视图比单视图提供更丰富的信息，现有MvC方法通常能取得令人满意的性能。然而在实际场景中，当视图存在噪声时，其性能可能严重退化。本文首先系统分析了噪声视图的缺陷，进而提出一种具有理论基础的深度MvC方法（即MvCAN）以解决该问题。具体地，我们提出一种新型MvC目标函数，通过允许跨视图的非共享参数与不一致聚类预测来降低噪声视图的副作用。此外，我们设计了一种非参数迭代过程，用于生成鲁棒的学习目标以挖掘多视图的有效信息。理论分析表明，MvCAN通过同时实现多视图一致性、互补性与噪声鲁棒性发挥作用。最后，在公开数据集上的实验证明，MvCAN优于现有最先进方法，并对噪声视图的存在具有鲁棒性。