Multiview Subspace Clustering Using Low-Rank Representation.

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Multiview subspace clustering methods leverage internal data structures but often fail to exploit intrinsic multiview characteristics.
• Existing approaches typically construct affinity matrices individually per view, neglecting cross-view correlations.

Purpose of the Study:

• To propose a novel Multiview Low-Rank Representation (MLRR) method for comprehensive multiview data correlation discovery.
• To enhance multiview subspace clustering by simultaneously exploiting diversity and consistency across multiple views.

Main Methods:

• Developed MLRR, which utilizes symmetric low-rank representations (LRRs) to exploit angular information of principal directions.
• Incorporated a diversity regularization term and a symmetry constraint on LRRs to capture view consistency and diversity.
• Employed convex optimization for efficient model calculation and a spectral clustering-based fusion strategy for compact representation.

Main Results:

• The MLRR model effectively captures angular information and enforces symmetry in LRRs.
• Experimental results on benchmark datasets demonstrate superior effectiveness and robustness compared to state-of-the-art algorithms.
• The proposed fusion strategy yields a compact, shared representation that captures intrinsic multiview data features.

Conclusions:

• MLRR offers a significant advancement in multiview subspace clustering by comprehensively exploiting multiview data correlations.
• The method's ability to handle diversity and consistency ensures robust and accurate clustering performance.
• MLRR provides an efficient and effective approach for analyzing complex multiview datasets.