A Neural Collapse Perspective on Feature Evolution in Graph Neural Networks

Keyphrases

• Graph Neural Network 100%
• Feature Evolution 100%
• Neural Collapse 100%
• Within-class 40%
• Mathematical Model 20%
• Image Classification 20%
• Classification Task 20%
• Graph Topology 20%
• Spectral Method 20%
• Graph Data 20%
• Symmetric Structure 20%
• Community Detection 20%
• Structural Conditions 20%
• Collapse Pattern 20%
• Training Error 20%
• Zero-training 20%
• Graph Features 20%
• Network Generalization 20%
• Partial Collapse 20%
• Feature Variability 20%
• Topology Evolution 20%
• Stochastic Block Model 20%
• Gradient Dynamics 20%
• Popular 20%
• Class Means 20%
• Class Feature 20%
• Deep Learning Classifiers 20%
• Heterophilic Graphs 20%

Computer Science

• Graph Neural Network 100%
• Image Classification 20%
• Topology Graph 20%
• Community Detection 20%
• Training Instance 20%
• Deep Feature 20%
• Training Error 20%
• Classification Task 20%
• Structured Data 20%
• Mathematical Modeling 20%
• Stochastic Block Model 20%
• Spectral Method 20%
• Theoretical model 20%

Mathematics

• Neural Network 100%
• Mathematical Modeling 20%
• Graph Model 20%
• Spectral Method 20%
• Structured Data 20%
• Topology 20%
• Stochastics 20%
• Community Detection 20%
• Model Block 20%