Embedding Regularizer Learning for Multi-View Semi-Supervised Classification.

Area of Science:

• Machine Learning
• Computer Vision
• Data Science

Background:

• Multi-view data classification with limited labels presents significant challenges.
• Existing methods struggle to effectively integrate information from diverse data sources.

Purpose of the Study:

• To propose an effective framework for multi-view semi-supervised classification using limited labeled data.
• To enhance classification performance by leveraging complementary information across multiple views.

Main Methods:

• Developed an embedding regularizer learning scheme for multi-view semi-supervised classification (ERL-MVSC).
• Integrated diversity, sparsity, and consensus principles into the framework.
• Utilized a linear regression model for view-specific regularizers and l2,1-norm for fused regularizer sparsity.
• Formulated as a joint optimization problem solved via coordinate descent.

Main Results:

• ERL-MVSC effectively incorporates complementary information from different views.
• The l2,1-norm regularization promotes sparse local structure, enhancing robustness to noise.
• Experimental results on real-world datasets demonstrate superior performance compared to existing algorithms.

Conclusions:

• The proposed ERL-MVSC framework offers a robust and effective solution for multi-view semi-supervised classification.
• The integration of diversity, sparsity, and consensus significantly improves classification accuracy with limited labels.