Two-step graph propagation for incomplete multi-view clustering.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Traditional clustering methods assume complete data, limiting their applicability.
• Existing incomplete multi-view clustering approaches often fail to capture high-order correlations and are computationally inefficient.
• Handling missing data in multi-view clustering remains a significant challenge.

Purpose of the Study:

• To propose a novel graph-based model for incomplete multi-view clustering.
• To effectively handle data incompleteness and capture high-order correlations across multiple views.
• To improve computational efficiency by decoupling the optimization procedure.

Main Methods:

• A graph-based model leveraging graph propagation to handle incomplete instances by translating them into incomplete graphs.
• Construction of a self-guided graph for global relationships and partial graphs for view-specific similarities.
• Low-rank tensor learning to capture high-order correlations across multiple views.
• A stepwise, decoupled optimization procedure for enhanced computational efficiency.

Main Results:

• The proposed graph propagation strategy effectively infers missing data entries, ensuring contextual relevance.
• The method successfully captures high-order correlations across multiple views using low-rank tensor learning.
• Experiments demonstrate superior performance and robustness compared to state-of-the-art methods, particularly with incomplete data.

Conclusions:

• The proposed graph propagation model offers a robust and efficient solution for incomplete multi-view clustering.
• The method effectively addresses data incompleteness and captures complex correlations, outperforming existing approaches.
• The decoupled optimization enhances efficiency, making the method practical for real-world applications.