View-shuffled clustering via the modified Hungarian algorithm.

Area of Science:

• Data Science
• Machine Learning
• Computer Vision

Background:

• Multi-view clustering typically requires accurate cross-view correspondence.
• Real-world data often violates this assumption, leading to the View-shuffled Problem (VsP).

Purpose of the Study:

• To propose a novel multi-view clustering method to address the VsP.
• To develop a method capable of handling data with unknown or partial cross-view correspondence.

Main Methods:

• Introduced View-shuffled Clustering via the Modified Hungarian Algorithm (VsC-mH).
• Employed global alignment and a modified Hungarian algorithm (mH) for intra-category alignment to establish cross-view correspondence.
• Utilized matrix factorization for data partitioning after alignment.
• Integrated alignment and partitioning for improved information interaction.

Main Results:

• VsC-mH effectively handles data with alignment ratios from 0% to 100%.
• Demonstrated convergence of the optimization algorithm through theoretical and experimental evidence.
• Achieved superior performance on six practical datasets compared to existing methods.

Conclusions:

• The proposed VsC-mH method offers a robust solution for multi-view clustering under the VsP.
• The integrated approach of alignment and partitioning enhances clustering quality.
• The method shows significant effectiveness and merits for real-world applications.