Relationship completion for incomplete multi-view clustering.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Real-world multi-view data often suffer from missing or incomplete samples.
• Existing Incomplete Multi-View Clustering (IMVC) methods struggle with utilizing structural information, maintaining data integrity, and handling clustering uncertainty.

Purpose of the Study:

• To propose a novel IMVC method that addresses limitations of existing approaches.
• To effectively handle missing values by leveraging structural relationships and ensuring data integrity.

Main Methods:

• A novel relationship completion module is designed to directly complete relationships of missing views.
• Attention layer fusion and high-confidence bootstrapping exploit multi-view complementary information.
• Semantic contrast learning and multi-view label distribution learning enhance consistent information extraction.

Main Results:

• The proposed method effectively addresses missing values while maintaining data integrity and mitigating error propagation.
• Experimental results on multiple real datasets demonstrate the superiority of the proposed method over state-of-the-art techniques.

Conclusions:

• The novel relationship completion module significantly improves IMVC performance.
• The method effectively handles incomplete multi-view data, offering a robust solution for clustering challenges.