TLRLF4MVC: Tensor Low-Rank and Low-Frequency for Scalable Multi-View Clustering

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Anchor-based multi-view clustering is effective for large datasets.
• Existing methods struggle with intra-view similarity or computational complexity.
• Efficient large-scale multi-view clustering remains a challenge.

Purpose of the Study:

• To develop a scalable multi-view clustering method that efficiently handles large datasets.
• To address limitations in current methods regarding intra-view similarity and computational cost.
• To improve clustering accuracy and efficiency in massive multi-view data.

Main Methods:

• Introduced a novel tensor low-frequency component (TLFC) operator for smooth sample representation.
• Integrated TLFC for intra-view similarity with tensor nuclear norm (TNN) and consensus regularization for inter-view correlations.
• Developed the tensor low-rank and low-frequency for scalable multi-view clustering (TLRLF4MVC) algorithm.
• Employed iterative optimization to balance intra-view similarity and inter-view complementary information.

Main Results:

• TLRLF4MVC significantly outperforms state-of-the-art methods on six large-scale multi-view datasets.
• The method demonstrates remarkable computational efficiency, especially for massive data.
• Learned embedding features are mapped into a smooth and compact subspace, enhancing clustering performance.

Conclusions:

• TLRLF4MVC offers a computationally efficient and accurate solution for large-scale multi-view clustering.
• The proposed TLFC operator effectively captures intra-view similarity.
• The integration of TNN and consensus regularization successfully exploits inter-view correlations.