Simplifying Scalable Subspace Clustering and Its Multi-View Extension by Anchor-to-Sample Kernel.

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Sparse subspace learning is crucial for spectral clustering but computationally expensive due to using complete sample dictionaries.
• Existing anchor-based methods, while improving scalability, often exhibit quadratic or cubic complexity concerning the number of anchors.

Purpose of the Study:

• To develop a more computationally efficient and scalable subspace clustering algorithm.
• To enhance the performance and robustness of multi-view clustering.

Main Methods:

• Derived a simplified problem to replace traditional scalable subspace clustering, achieving linear complexity with respect to both samples and anchors.
• Introduced a separate fusion strategy for multi-view extensions, improving inter-view difference measurement and avoiding alternate optimization.

Main Results:

• The proposed method significantly reduces time overhead compared to existing approaches.
• Demonstrated superior performance in clustering tasks, particularly for multi-view scenarios.

Conclusions:

• The new formulation offers enhanced scalability and efficiency for subspace clustering.
• The separate fusion strategy leads to more robust and effective multi-view clustering.