Soft label collaborative view consistency enhancement with application to incomplete multi-view clustering.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Incomplete Multi-View Clustering (IMVC) methods face challenges with inaccurate data imputation and degraded feature extraction from low-quality views.
• Existing approaches often struggle to effectively handle missing data, impacting overall clustering performance.

Purpose of the Study:

• To propose a novel IMVC framework, Soft Label Collaborative View Consistency Enhancement (SLC_CE), to address the limitations of existing methods.
• To enhance feature embeddings and improve data imputation accuracy in IMVC tasks.

Main Methods:

• Leveraging Transformer encoders for a soft-label view information interaction module to boost feature embeddings.
• Employing soft labels for collaborative imputation of missing features to handle incomplete multi-view data.
• Implementing a multi-level consistency enhancement strategy across features and soft labels for robust extraction and imputation.

Main Results:

• The proposed SLC_CE method demonstrates superior performance compared to state-of-the-art methods on benchmark datasets.
• Effective enhancement of view feature embeddings and accurate imputation of missing data were achieved.
• High-quality feature extraction and imputation were ensured through the consistency enhancement strategy.

Conclusions:

• The SLC_CE framework offers a significant advancement in tackling incomplete multi-view clustering challenges.
• The method provides a robust solution for enhancing clustering performance with incomplete multi-view data.
• Experimental results validate the effectiveness and superiority of SLC_CE in real-world IMVC applications.