Low-rank tensor constrained co-regularized multi-view spectral clustering.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Graph-oriented clustering methods excel at analyzing multi-view data but often treat all views equally.
• Existing methods neglect spatial and complementary information by processing features independently.
• These limitations hinder the effective exploitation of complex structures within multi-view datasets.

Purpose of the Study:

• To develop an enhanced multi-view spectral clustering model addressing limitations of current graph-based methods.
• To improve clustering robustness and accuracy by adaptively weighting individual data views.
• To better leverage high-order and complementary information inherent in multi-view data.

Main Methods:

• Proposed a novel model employing a weighted tensor nuclear norm to characterize consistency among indicator matrices.
• The weighted tensor nuclear norm explicitly captures salient differences in singular values.
• Implemented adaptive view weighting to enhance the algorithm's robustness.

Main Results:

• The enhanced model effectively exploits high-order and complementary information.
• Demonstrated improved mining of consistency between indicator matrices.
• Extensive experiments confirmed the method's efficiency and superior performance.

Conclusions:

• The proposed weighted tensor nuclear norm effectively addresses limitations in existing multi-view spectral clustering.
• Adaptive view weighting significantly enhances model robustness and clustering performance.
• The method offers a powerful approach for analyzing complex multi-view data structures.