SPGMVC: Multiview Clustering via Partitioning the Signed Prototype Graph.

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Multiview clustering (MVC) leverages multiple data perspectives for improved performance.
• Existing MVC methods face computational challenges with large-scale datasets.
• Current binary MVC (BMVC) approaches neglect geometric information, impacting clustering accuracy.

Purpose of the Study:

• To develop an efficient and accurate framework for large-scale multiview clustering.
• To address the limitations of existing BMVC methods, particularly regarding geometric information and parameter selection.
• To introduce a unified framework integrating consensus binary coding, code compression, and signed prototype graph partitioning.

Main Methods:

• Proposed the Signed Prototype Graph Multiview Clustering (SPGMVC) framework.
• Integrated consensus binary coding and code compression (CC) for dimensionality reduction and efficiency.
• Utilized signed graph (SG) partitioning based on edge relationships and an alternating minimization strategy for optimization.
• Developed an automatic parameter selection strategy to simplify model tuning.

Main Results:

• SPGMVC effectively captures underlying data structures through signed graph partitioning, enhancing clustering accuracy (ACC).
• Code compression (CC) reduces graph scale, improving computational efficiency for large datasets.
• The alternating minimization strategy achieves nearly linear time and space complexity, suitable for large-scale tasks.
• Automatic parameter selection eliminates the need for manual tuning, simplifying the process.

Conclusions:

• SPGMVC provides a unified, efficient, and accurate solution for large-scale multiview clustering.
• The method overcomes limitations of previous BMVC approaches by incorporating geometric information and simplifying parameter selection.
• SPGMVC demonstrates superior performance in comprehensive experiments, offering a practical advancement in the field.