Accelerated low-rank representation for subspace clustering and semi-supervised classification on large-scale data.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Scalability of Low-Rank Representation (LRR) to large datasets is hindered by computationally intensive Singular Value Decomposition (SVD) iterations.
• Existing methods to accelerate LRR are often computationally heavy and can degrade representation quality.

Purpose of the Study:

• To propose a novel and efficient method, Accelerated Low-Rank Representation (ALRR), for handling large-scale data.
• To improve the computational efficiency and accuracy of low-rank representation.

Main Methods:

• ALRR integrates matrix factorization with nuclear-norm minimization.
• Transforms large coefficient matrices into smaller ones for efficient SVD computation.
• Optimization complexity is linear with the number of data points.

Main Results:

• ALRR demonstrates convexity, accuracy, robustness, and efficiency on large-scale data.
• Comparative studies on real image datasets for subspace clustering and semi-supervised classification show ALRR's superiority.
• The size of the transformed matrix is independent of the number of data points.

Conclusions:

• ALRR offers a significant advancement in scalable low-rank representation.
• The method effectively addresses the computational challenges of LRR for large datasets.
• ALRR proves superior to state-of-the-art methods in tested applications.