Multi-View Clustering Via Bilaterally Constrained Anchor Graph.

Area of Science:

• Data Science
• Machine Learning
• Computer Vision

Background:

• Anchor similarity matrices are crucial for efficient clustering but suffer from row-column imbalance.
• Existing methods often lack constraints on matrix columns, hindering comprehensive data structure representation.

Purpose of the Study:

• To propose a novel method for imposing meaningful constraints on anchor similarity matrix columns.
• To develop a multi-view clustering framework with bilateral constraints for balanced data representation.

Main Methods:

• Introduced Multi-view Clustering via Bilaterally constrained anchor Graph (MCBG).
• Developed a fused anchor similarity matrix with distinct row and column constraints.
• Incorporated a rank constraint on the Laplacian matrix for a post-processing-free framework.
• Utilized an efficient alternating iterative optimization algorithm.

Main Results:

• MCBG achieved a balanced and expressive anchor similarity distribution, avoiding degenerate cases.
• The proposed method demonstrated superior performance in multi-view clustering tasks.
• Experimental validation confirmed the effectiveness of the bilateral constraint approach.

Conclusions:

• MCBG offers a robust solution to the row-column imbalance in anchor similarity matrices.
• The bilateral constraint strategy enhances multi-view clustering accuracy and efficiency.
• This work provides a significant advancement in graph-based clustering methodologies.