Semi-supervised learning for multi-view and non-graph data using Graph Convolutional Networks.

Area of Science:

• Machine Learning
• Computer Vision
• Data Science

Background:

• Semi-supervised learning is crucial for handling large unlabeled datasets.
• Graph Convolutional Networks (GCNs) are effective for graph-structured data but limited for non-graph data.
• A gap exists in applying GCNs to multi-view, non-graph data like image collections.

Purpose of the Study:

• To develop a novel deep semi-supervised multi-view classification model for non-graph data.
• To bridge the gap between GCNs and multi-view image classification.
• To improve classification accuracy in scenarios with limited labeled data.

Main Methods:

• Developed a deep semi-supervised multi-view classification model.
• Independently reconstructed graphs for each data view using a semi-supervised method.
• Adaptively merged individual graphs into a unified consensus graph.
• Employed a unified GCN framework with a label smoothing constraint on the consensus graph.

Main Results:

• The proposed model demonstrated superior performance in both graph generation and classification.
• Experimental results on seven multi-view image datasets showed consistent outperformance.
• The model surpassed traditional GCNs and other existing semi-supervised multi-view classification methods.

Conclusions:

• The novel model effectively addresses the challenge of applying GCNs to non-graph multi-view data.
• The approach offers a significant advancement in semi-supervised classification for image datasets.
• The method provides a robust solution for scenarios where data labeling is expensive.