Sparse subspace clustering for data with missing entries and high-rank matrix completion.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Subspace clustering methods struggle with incomplete datasets due to missing entries.
• Existing matrix completion techniques often fail with high-rank matrices common in practical scenarios.

Purpose of the Study:

• To propose a novel algorithm for subspace clustering with incomplete data.
• To address the challenge of high-rank matrix completion in the context of subspace clustering.

Main Methods:

• Developed a Sparse Representation with Missing Entries and Matrix Completion (SRM²C) algorithm.
• The algorithm iteratively computes sparse representation coefficients and imputes missing data.
• Minimizes representation coefficients, errors, and matrix rank for data recovery.

Main Results:

• The proposed SRM²C algorithm demonstrates superior performance in both subspace clustering and matrix completion.
• Experiments on synthetic and natural image data validate the algorithm's effectiveness.
• Outperforms existing methods in handling incomplete data and high-rank matrices.

Conclusions:

• The novel SRM²C algorithm offers an effective solution for subspace clustering with incomplete data.
• The method successfully addresses limitations of conventional matrix completion for high-rank matrices.
• Provides a robust approach for real-world applications involving missing data in clustering.