Mask-informed Deep Contrastive Incomplete Multi-view Clustering

TL;DR

Mask-IMvC tackles incomplete multi-view clustering by eliminating missing-view imputation and instead using a mask-informed fusion to form a discriminative view-common representation $\mathbf{F}$. It introduces a prior knowledge-assisted contrastive learning framework that injects neighborhood information across views via a re-weighted contrastive loss $\zeta_{wcl}$ and a QR-based clustering projection on $\mathbf{Y}$. The method achieves state-of-the-art results on four benchmark datasets in both complete and incomplete settings and is validated by extensive ablations showing the contributions of mask-informed fusion and PKCL. This imputation-free approach improves robustness to missing data and offers a scalable solution for real-world multi-view clustering applications.

Abstract

Multi-view clustering (MvC) utilizes information from multiple views to uncover the underlying structures of data. Despite significant advancements in MvC, mitigating the impact of missing samples in specific views on the integration of knowledge from different views remains a critical challenge. This paper proposes a novel Mask-informed Deep Contrastive Incomplete Multi-view Clustering (Mask-IMvC) method, which elegantly identifies a view-common representation for clustering. Specifically, we introduce a mask-informed fusion network that aggregates incomplete multi-view information while considering the observation status of samples across various views as a mask, thereby reducing the adverse effects of missing values. Additionally, we design a prior knowledge-assisted contrastive learning loss that boosts the representation capability of the aggregated view-common representation by injecting neighborhood information of samples from different views. Finally, extensive experiments are conducted to demonstrate the superiority of the proposed Mask-IMvC method over state-of-the-art approaches across multiple MvC datasets, both in complete and incomplete scenarios.

Figures (8)

• Figure 1: Comparison of two different IMvC paradigms. (a) The imputation-based IMvC. (b) Our proposed mask-informed deep contrastive IMvC.
• Figure 2: Illustration of mask-informed deep contrastive incomplete multi-view clustering (Mask-IMvC). The view-complete parts of IMvC data are first processed through their encoders to extract view-specific latent features. Next, a mask-informed fusion module aggregates the representations into a unified view-common one, which is then used to reconstruct the view complete parts of IMvC data via view-specific decoders. Finally, the prior knowledge from different views is fused via the mask-informed fusion strategy to assist the contrastive learning on the view-common representation.
• Figure 3: The parameter sensitivity of the proposed method on four multi-view datasets in terms of ACC, NMI, ARI, and Fscore, respectively.
• Figure 4: The convergence analysis of the proposed method on four benchmark multi-view datasets in terms of ACC, NMI, ARI, and Fscore, respectively.
• Figure 5: The visualization results of the proposed method under 1, 50, 100 and 200 epochs on CUB dataset.