Multiview Clustering via Block Diagonal Graph Filtering.

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Graph-based multiview clustering methods leverage feature and topological information for improved accuracy.
• Existing methods often use predetermined graph filters, limiting adaptability to clustering tasks.
• Poor separability of filtered features in current approaches hinders effective clustering.

Purpose of the Study:

• To propose a novel multiview clustering method that learns cluster-friendly graph filters.
• To address limitations of predetermined graph filters and poor feature separability in existing methods.
• To develop a unified framework integrating graph filter learning and consensus graph acquisition.

Main Methods:

• Introduced Multiview Clustering via Block Diagonal Graph Filtering (MvC-BDGF).
• Designed a block diagonal graph filter with localized characteristics for discriminating features.
• Developed a unified framework for simultaneous learning of optimal filters and clustering labels.
• Employed an iterative solver based on the coordinate descent method for optimization.

Main Results:

• The proposed MvC-BDGF model effectively learns cluster-friendly graph filters.
• Integrated filter learning with consensus graph acquisition for optimal clustering.
• Achieved superior clustering performance on benchmark datasets.
• Demonstrated the effectiveness and superiority of the MvC-BDGF model through extensive experiments.

Conclusions:

• MvC-BDGF offers a significant advancement in multiview clustering by learning adaptive graph filters.
• The unified framework enhances feature discrimination and clustering accuracy.
• The method provides a robust solution for complex clustering tasks, outperforming existing approaches.