Rank Consistency Induced Multiview Subspace Clustering via Low-Rank Matrix Factorization

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Multiview subspace clustering leverages complementary information from multiple data sources.
• Existing methods often rely on a shared self-expressiveness coefficient matrix, limiting view-specific structural learning.
• Exploiting view-specific information is crucial for robust multiview clustering.

Purpose of the Study:

• To propose a novel multiview subspace clustering model that enforces rank consistency among view-specific self-expressiveness coefficient matrices.
• To develop a model that simultaneously learns a consistent subspace structure and exploits complementary information.
• To address the limitations of shared coefficient matrices in existing approaches.

Main Methods:

• A rank consistency induced multiview subspace clustering model is proposed.
• Low-rank structure is parameterized using tri-factorization with orthogonal constraints.
• A nonconvex optimization problem is formulated and solved with an efficient algorithm guaranteeing convergence.

Main Results:

• The proposed model achieves consistent low-rank structures across multiple views.
• It effectively exploits complementary information from view-specific self-expressiveness matrices.
• Extensive experiments show superior performance compared to state-of-the-art multiview clustering methods.

Conclusions:

• The rank consistency approach enhances multiview subspace clustering by promoting structural consistency.
• The proposed model effectively integrates view-specific information for improved clustering.
• The developed optimization algorithm ensures practical applicability and convergence.