Learning multi-level topology representation for multi-view clustering with deep non-negative matrix factorization.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Multi-view clustering aims to group objects while preserving cluster structures across all data views.
• Existing algorithms struggle to effectively balance the diversity and consistency inherent in multi-view data.
• This limitation leads to suboptimal performance in characterizing shared and specific components across views.

Purpose of the Study:

• To propose a novel Multi-View Clustering with Deep non-negative matrix factorization and Multi-Level Representation (MVC-DMLR) learning algorithm.
• To address the limitations of current methods in balancing diversity and consistency in multi-view data.
• To integrate feature learning, multi-level topology representation, and clustering within a unified framework.

Main Methods:

• Employs deep non-negative matrix factorization (DNMF) to learn multi-level representations (deep features) of objects.
• Constructs multi-level graphs for each view from these representations to capture relationships at various resolutions.
• Formulates the integrated learning of multi-level representation, topology, and clustering as an optimization problem.

Main Results:

• Experimental results demonstrate the effectiveness of MVC-DMLR.
• The proposed method significantly outperforms baseline algorithms.
• Superiority is evidenced by higher accuracy, F1-score, normalized mutual information, and adjusted rand index.

Conclusions:

• MVC-DMLR offers a superior approach to multi-view clustering by effectively managing diversity and consistency.
• The integration of deep feature learning and multi-level topology representation is key to its enhanced performance.
• This method provides a robust framework for analyzing complex multi-view datasets.