Accelerated Stochastic Variance Reduction Gradient Algorithms for Robust Subspace Clustering.

Area of Science:

• Computer Vision
• Machine Learning
• Data Science

Background:

• Robust face clustering is crucial for applications like security and surveillance.
• Existing algorithms struggle with noisy data, such as occluded faces.
• Deterministic subspace clustering methods face computational challenges for large datasets.

Purpose of the Study:

• To propose an efficient algorithm for robust subspace clustering, particularly for face data.
• To enhance the performance of subspace clustering in the presence of noise and occlusions.
• To address the high computational complexity of existing deterministic methods.

Main Methods:

• Development of the first accelerated stochastic variance reduction gradient (RASVRG) algorithm for robust subspace clustering.
• Introduction of a novel momentum acceleration technique integrated into the RASVRG algorithm.
• Evaluation using real-world face datasets with varying levels of pixel corruption and occlusion.

Main Results:

• The proposed RASVRG algorithm demonstrates superior accuracy and robustness compared to state-of-the-art methods, especially with noisy and occluded face data.
• The momentum acceleration technique enhances convergence rate and practical efficiency for both strongly and not strongly convex models.
• RASVRG achieved better performance in accuracy across various experimental setups.

Conclusions:

• The RASVRG algorithm offers a significant advancement in robust face clustering, overcoming limitations of previous methods.
• The algorithm provides a computationally efficient and accurate solution for large-scale face clustering tasks.
• RASVRG shows strong potential for real-world applications in security, surveillance, and embedded systems.