Scaled Simplex Representation for Subspace Clustering.

Area of Science:

• Data Science
• Machine Learning
• Computer Vision

Background:

• Subspace clustering (SC) methods leverage data's self-expressive property for effective clustering.
• Existing SC methods often lose data correlations due to transformations of negative coefficients in affinity matrix construction.
• The affine constraint in traditional SC methods lacks flexibility for real-world applications.

Purpose of the Study:

• To introduce a novel Scaled Simplex Representation (SSR) for subspace clustering.
• To address limitations of existing SC methods concerning negative coefficients and affine constraints.
• To develop a more accurate and efficient subspace clustering algorithm.

Main Methods:

• Proposed a Scaled Simplex Representation (SSR) incorporating non-negative constraints for physically meaningful coefficients.
• Constrained coefficient vectors to sum to a scalar for enhanced discrimination.
• Reformulated the SSR-based SC (SSRSC) model as a linear equality-constrained problem.
• Employed the alternating direction method of multipliers (ADMM) for efficient model solving.

Main Results:

• The proposed SSRSC algorithm demonstrates high efficiency in experiments.
• SSRSC significantly outperforms existing state-of-the-art SC methods in terms of accuracy.
• Experimental results validate the effectiveness of the non-negative and scalar sum constraints.

Conclusions:

• The Scaled Simplex Representation (SSR) offers a superior approach to subspace clustering.
• SSRSC provides a more robust and accurate clustering solution compared to traditional methods.
• The developed method is efficient and applicable to benchmark datasets.