Low-Rank Graph Completion-Based Incomplete Multiview Clustering.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Incomplete multiview clustering (IMVC) is crucial for handling datasets with missing data across different views.
• Existing graph-based IMVC methods often fail to capture cross-view relationships and global structure information.
• Current methods struggle to adaptively recover incomplete data structures using both global and cross-view information.

Purpose of the Study:

• To propose a unified optimization framework for IMVC that addresses limitations of existing methods.
• To develop a novel approach that effectively explores potential relationships among views and exploits global structure.
• To adaptively recover incomplete graph structures and obtain complete affinity relationships for improved clustering.

Main Methods:

• Introduced adaptive graph embedding to explore potential relationships among views.
• Incorporated a low-rank constraint to effectively exploit global structure information.
• Unified intra-view, cross-view, and global information for adaptive graph structure recovery.

Main Results:

• The proposed low-rank graph completion-based IMVC (LRGR_IMVC) method demonstrates superior clustering performance.
• Experimental results show significant improvements over existing state-of-the-art methods on multiple datasets.
• The method effectively handles missing data by adaptively recovering incomplete affinity relationships.

Conclusions:

• The LRGR_IMVC framework offers a robust solution for incomplete multiview clustering.
• The integration of adaptive graph embedding and low-rank constraints enhances clustering accuracy.
• This approach provides a significant advancement in machine learning for handling complex, incomplete data.