Nonlinear multi-view clustering for non-negative matrix factorization.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Deep multi-view clustering methods offer advanced learning capabilities but face challenges with opaque processes, stability issues, and limited nonlinear fitting.
• Existing Non-negative Matrix Factorization (NMF) methods are often constrained by narrow parameter spaces and reduced robustness.

Purpose of the Study:

• To develop a robust and interpretable multi-view clustering framework that addresses limitations of current deep learning and NMF approaches.
• To enhance the nonlinear fitting capabilities and stability of deep clustering models.

Main Methods:

• Proposed a nonlinear Non-negative Matrix Factorization (NMF) multi-view clustering framework.
• Integrated traditional NMF optimization principles into a deep model for improved interpretability and stability.
• Employed partially parameterized NMF iterations and a cross-view contrastive loss for enhanced nonlinear fitting, expanded parameter space, and inter-view diversity learning.

Main Results:

• The proposed framework demonstrated enhanced stability and robustness compared to existing methods.
• Experiments showed superior performance of the new method over state-of-the-art multi-view clustering techniques on multiple datasets.
• The integration of NMF principles improved the interpretability of the deep clustering model.

Conclusions:

• The developed nonlinear NMF multi-view clustering framework effectively addresses key limitations in current deep clustering methods.
• The approach offers a more robust, stable, and interpretable solution for multi-view clustering tasks.
• Future work could explore further enhancements in nonlinear fitting and cross-view information integration.