Efficient Multi-View Clustering via Unified and Discrete Bipartite Graph Learning.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Existing graph-based multi-view clustering (MVC) algorithms face challenges with high computational complexity, limited graph learning strategies, and reliance on k-means for discretization.
• These limitations hinder their application in large-scale datasets and direct learning of discrete cluster structures.

Purpose of the Study:

• To propose an efficient multi-view clustering (MVC) approach that addresses the limitations of existing graph-based methods.
• To introduce Unified and Discrete Bipartite Graph Learning (UDBGL) for improved performance and scalability.

Main Methods:

• UDBGL incorporates anchor-based subspace learning for view-specific bipartite graph creation.
• It employs bipartite graph fusion with adaptive weights to learn a view-consensus bipartite graph.
• A Laplacian rank constraint ensures discrete cluster structures, and a unified objective function enables simultaneous learning.

Main Results:

• The proposed UDBGL method achieves discrete clustering directly, eliminating the need for post-processing partitioning.
• UDBGL demonstrates linear time complexity concerning data size, enhancing efficiency for large-scale scenarios.
• Experimental results on diverse multi-view datasets confirm the robustness and efficiency of the UDBGL approach.

Conclusions:

• UDBGL offers an efficient and robust solution for multi-view clustering problems.
• The unified framework and discrete learning capability of UDBGL overcome key limitations of prior methods.
• The approach is suitable for large-scale applications due to its linear time complexity.