Learnable Graph Convolutional Network With Semisupervised Graph Information Bottleneck.

Area of Science:

• Machine Learning
• Graph Neural Networks

Background:

• Graph Convolutional Networks (GCNs) are widely used for semisupervised classification.
• Existing GCN methods often use fixed graphs, limiting their ability to capture dynamic local and global relationships and handle noisy data.

Purpose of the Study:

• To propose a learnable GCN framework that dynamically optimizes graph structures.
• To enhance semisupervised classification by jointly learning graph structure and feature propagation.

Main Methods:

• A unified network integrating graph learning and feature propagation.
• Dual-GCN-based meta-channels to explore local and global relations.
• A semisupervised graph information bottleneck (SGIB) for graph structural learning (GSL) and minimal sufficient representations.

Main Results:

• The proposed model effectively learns optimal graph structures.
• Dual-GCN meta-channels successfully capture both local and global relationships.
• SGIB minimizes noisy data interference and enhances representation learning.

Conclusions:

• The novel GCN framework demonstrates robustness and superior performance over state-of-the-art fixed-structure graph methods.
• Dynamic graph learning significantly improves semisupervised node classification accuracy.