Multiview Latent Space Learning With Feature Redundancy Minimization

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Multiview learning shows promise but faces challenges with feature redundancy.
• Existing methods struggle to balance complementary information and view-specific properties, leading to performance degradation.

Purpose of the Study:

• To propose a novel multiview latent space learning framework.
• To address feature redundancy and improve the exploitation of complementary information across multiple views.

Main Methods:

• Projecting multi-view features onto a latent space.
• Learning shared and view-specific dictionaries to capture correlations and preserve properties.
• Utilizing the Hilbert-Schmidt independence criterion for diversity and local structure preservation.

Main Results:

• The proposed framework effectively mitigates feature redundancy.
• Demonstrated superior performance compared to state-of-the-art methods on six public datasets.
• Successfully balanced complementary information exploitation and view-specific property preservation.

Conclusions:

• The novel framework offers an effective solution for multiview learning challenges.
• Highlights the importance of feature redundancy minimization and diversity constraints.
• Provides a robust approach for enhancing learning performance in multiview scenarios.