Uniform Distribution Non-Negative Matrix Factorization for Multiview Clustering.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Multiview data processing enhances clustering by integrating information from diverse sources.
• Non-negative matrix factorization (NMF) is a common technique for dimensionality reduction in multiview data.
• Clustering performance is sensitive to data distribution within the learned subspace.

Purpose of the Study:

• To propose a novel tri-factorization-based NMF model incorporating an embedding matrix.
• To improve clustering performance by learning discriminative representations with uniform data distribution.
• To develop a gradient-based algorithm for solving the proposed model and analyze its properties.

Main Methods:

• A tri-factorization NMF model with an embedding matrix is developed.
• A new lemma for partial derivation of the trace function is proposed and proven.
• A gradient-based optimization algorithm is designed to solve the model.

Main Results:

• The proposed model generates decompositions with uniform distribution, leading to more discriminative learned representations.
• The resulting consensus matrix better represents multiview data in the subspace, improving clustering accuracy.
• Experimental results on six real-world datasets demonstrate the superiority of the proposed algorithm over baseline methods.

Conclusions:

• The proposed tri-factorization NMF model effectively enhances multiview data clustering.
• The embedding matrix and uniform distribution objective contribute to improved representation learning.
• The developed gradient-based algorithm is efficient and converges well, validated by empirical evidence.