Consensus Kernel K-Means Clustering for Incomplete Multiview Data.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Multiview clustering integrates information from multiple data sources to improve clustering accuracy.
• Existing methods struggle with incomplete data across different views.
• Prior work unified multiview clustering and imputation but neglected between-view consistency.

Purpose of the Study:

• To propose a unified learning framework for incomplete multiview clustering.
• To address the limitations of existing methods by incorporating between-view consistency.
• To simultaneously impute missing data and learn a consistent clustering result.

Main Methods:

• A novel unified learning method is proposed for incomplete multiview clustering.
• The method explicitly models between-view consistency by measuring similarity between individual view clusters and a consensus cluster.
• Incomplete views are imputed to optimize clustering within each view while preserving inter-view consistency.

Main Results:

• The proposed method demonstrates superior performance on both synthetic and real-world incomplete multiview datasets.
• Experimental comparisons show significant improvements over state-of-the-art methods.
• The explicit modeling of between-view consistency is shown to be crucial for enhanced performance.

Conclusions:

• The proposed unified learning method effectively handles incomplete multiview clustering by integrating imputation and consistency modeling.
• The approach offers a significant advancement in multiview clustering research, particularly for datasets with missing information.
• The findings highlight the importance of inter-view consistency for robust multiview clustering.