Dual-Correlation-Guided Anchor Learning for Scalable Incomplete Multi-View Clustering.

Area of Science:

• Computer Science
• Machine Learning
• Data Mining

Background:

• Incomplete multi-view clustering (IMC) faces challenges in learning effective representations from heterogeneous data.
• Existing anchor-based IMC methods struggle with unstable anchor generation and imbalanced anchor capabilities across views.

Purpose of the Study:

• To propose a novel Dual-Correlation-Guided Anchor Learning (DCGA) method for scalable and efficient IMC.
• To address deficiencies in anchor generation and inter-view anchor coherence in existing IMC models.

Main Methods:

• DCGA learns informative anchor spaces by integrating intra-view and inter-view correlations.
• An Anchor-as-a-Bottleneck (A3B) strategy stabilizes intra-view anchor spaces using Information Bottleneck (IB) principles.
• An Informative Anchor Constraint (IAC) aligns anchor spaces across different views.

Main Results:

• DCGA demonstrates superior effectiveness and efficiency compared to 11 state-of-the-art IMC methods across seven datasets.
• The method successfully generates stable and informative anchors, improving clustering accuracy.
• DCGA effectively handles data redundancy and preserves essential information from each view.

Conclusions:

• DCGA offers a robust and scalable solution for incomplete multi-view clustering.
• The proposed method enhances representation learning by effectively utilizing multi-view correlations.
• DCGA represents a significant advancement in unsupervised anchor learning for IMC.