Quality-Aware Robust Multi-View Clustering for Heterogeneous Observation Noise

TL;DR

This work proposes a novel framework termed Quality-Aware Robust Multi-View Clustering (QARMVC), which employs an information bottleneck mechanism to extract intrinsic semantics for view reconstruction and utilizes the resulting reconstruction discrepancy to precisely quantify fine-grained contamination intensity and derive instance-level quality scores.

Abstract

Deep multi-view clustering has achieved remarkable progress but remains vulnerable to complex noise in real-world applications. Existing noisy robust methods predominantly rely on a simplified binary assumption, treating data as either perfectly clean or completely corrupted. This overlooks the prevalent existence of heterogeneous observation noise, where contamination intensity varies continuously across data. To bridge this gap, we propose a novel framework termed Quality-Aware Robust Multi-View Clustering (QARMVC). Specifically, QARMVC employs an information bottleneck mechanism to extract intrinsic semantics for view reconstruction. Leveraging the insight that noise disrupts semantic integrity and impedes reconstruction, we utilize the resulting reconstruction discrepancy to precisely quantify fine-grained contamination intensity and derive instance-level quality scores. These scores are integrated into a hierarchical learning strategy: at the feature level, a quality-weighted contrastive objective is designed to adaptively suppress the propagation of noise; at the fusion level, a high-quality global consensus is constructed via quality-weighted aggregation, which is subsequently utilized to align and rectify local views via mutual information maximization. Extensive experiments on five benchmark datasets demonstrate that QARMVC consistently outperforms state-of-the-art baselines, particularly in scenarios with heterogeneous noise intensities.

Figures (5)

• Figure 1: An illustrative diagram of heterogeneous noise intensity in a multi-view scenario. The diagram displays Camera, LiDAR, and Audio views under varying environmental conditions. From left to right, the data quality exhibits a continuous degradation process from clean to severe noise, rather than a simple binary state.
• Figure 2: The framework consists of four modules: (a) quality score estimation, where an information bottleneck mechanism quantifies heterogeneous noise intensity to derive instance-view specific quality scores; (b) quality-aware representation learning, which utilizes these scores to re-weight contrastive learning, thereby suppressing noisy anchors; (c) global alignment, where local views are aligned with a robust high-quality global consensus via mutual information maximization; (d) cluster assignment, which imposes deep divergence clustering loss on the global representation to optimize the cluster structure and classifier head.
• Figure 3: Noise score analysis on the ALOI dataset.
• Figure 4: Sensitivity analysis on MNIST-USPS dataset with 10% Noise.
• Figure 5: Visualization on MNIST-USPS dataset with 10% Noise.