Multiview Subspace Clustering via Low-Rank Symmetric Affinity Graph

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Multiview subspace clustering (MVSC) analyzes datasets with multiple views, but existing methods often overlook crucial consistent and angular information.
• This limitation hinders the comprehensive understanding of complex data structures.

Purpose of the Study:

• To propose a novel MVSC method, Low-Rank Symmetric Graph Clustering (LSGMC), that effectively utilizes both consistent and angular information from different data views.
• To enhance the accuracy and robustness of multiview clustering by addressing limitations in existing approaches.

Main Methods:

• LSGMC decomposes the coefficient matrix into three factors to capture consistent low-rank structures across views.
• It incorporates a symmetry constraint for weight consistency and a fusion mechanism for inherent data structure analysis.
• The Schatten p-norm is employed for low-rank approximation, and an adaptive information reduction strategy generates a high-quality similarity matrix.

Main Results:

• LSGMC demonstrated superior clustering performance across 11 diverse datasets.
• The proposed method outperformed ten state-of-the-art multiview clustering techniques in experimental evaluations.
• The integration of consistent and angular information proved effective in improving clustering outcomes.

Conclusions:

• LSGMC offers a significant advancement in multiview subspace clustering by effectively leveraging complementary information from multiple views.
• The method's ability to handle noise and redundancy, coupled with its superior performance, makes it a valuable tool for complex data analysis.