Multi-view clustering via multi-manifold regularized non-negative matrix factorization.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Non-negative matrix factorization (NMF) is competitive for multi-view clustering.
• Standard NMF methods struggle to preserve local data geometry.
• Existing NMF-based multi-view clustering lacks robust local structure preservation.

Purpose of the Study:

• To propose a novel multi-manifold regularized non-negative matrix factorization (MMNMF) framework.
• To enhance multi-view clustering by preserving the local geometrical structure of data manifolds.
• To develop a flexible framework with multiple instances for improved clustering performance.

Main Methods:

• Developed a multi-manifold regularized non-negative matrix factorization (MMNMF) framework.
• Incorporated consensus manifold and consensus coefficient matrix.
• Introduced multi-manifold regularization to preserve local geometry.
• Created four framework instances by varying consensus manifold and coefficient matrix construction methods.

Main Results:

• The proposed MMNMF algorithms effectively preserve the locally geometrical structure of multi-view data.
• Experimental results demonstrate superior performance compared to existing NMF-based multi-view clustering algorithms.
• The framework's flexibility allows for tailored approaches to multi-view clustering problems.

Conclusions:

• MMNMF offers a significant advancement in multi-view clustering by integrating manifold regularization.
• The proposed framework effectively addresses the limitations of traditional NMF in preserving local data structures.
• MMNMF provides a robust and high-performing solution for complex multi-view clustering tasks.