Large-Scale Multiview Clustering via Joint Learning of Anchor Representation and Multigraph Alignment

Area of Science:

• Data Science
• Machine Learning
• Computer Vision

Background:

• Anchor-based clustering is a key technique for large-scale data.
• Multiview data presents challenges in balancing individual anchor graph distinctiveness with overall consistency.

Purpose of the Study:

• To propose a large-scale multiview clustering (MVC) method, ARMGA, that jointly learns anchor representation and multigraph alignment.
• To address the challenge of balancing distinctiveness and consistency in multiview anchor-based clustering.

Main Methods:

• ARMGA utilizes a unified framework for concurrent learning of single-view anchor representations and virtual graph-based multigraph alignment.
• It employs the Schatten-p norm on a tensor for adaptive anchor representation to reinforce cross-view consistency.
• Cosine angle information from low-rank representation is used to attenuate noise and reduce computational complexity.

Main Results:

• ARMGA demonstrated significant improvements in clustering performance, with a 2%-10% increase over other algorithms on nine datasets.
• The method maintained lower time complexity compared to existing approaches.
• ARMGA effectively leverages complementary information across views to enhance overall structure and consensus.

Conclusions:

• ARMGA offers an effective solution for large-scale multiview clustering by integrating anchor representation learning and multigraph alignment.
• The proposed method enhances cross-view consistency and robustness to noise.
• ARMGA achieves superior clustering performance with improved efficiency.