Graph Convolutional Network With Self-Augmented Weights for Semi-Supervised Multi-View Learning.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computer Vision

Background:

• Graph-based deep learning excels in multi-view tasks by leveraging data interconnections.
• Existing methods often discard less important views or fail to enhance key views through simple weighting strategies.

Purpose of the Study:

• To propose a novel graph convolutional network (GCN) with a self-augmented weight strategy for semi-supervised multi-view learning.
• To address the limitation of existing methods that ignore complementary information from less important views.

Main Methods:

• A self-augmented weight strategy based on exponential series integration is developed to preserve and enhance view importance.
• An orthogonal constraint layer with forced orthogonal weights is introduced to improve representation discriminability.

Main Results:

• The proposed self-augmented weight strategy adaptively assigns non-zero weights to preserve complementary information and higher weights to key views.
• The orthogonal constraint layer enhances the distinctiveness of learned representations.
• Extensive experiments confirm the superior performance of the proposed method.

Conclusions:

• The novel GCN with self-augmented weights effectively fuses multi-view information by preserving complementary data.
• The method offers a significant improvement over existing approaches in semi-supervised multi-view learning tasks.