Oversquashing in GNNs through the lens of information contraction and graph expansion

Keyphrases

• Graph Neural Network 100%
• Expander Graphs 100%
• Contraction-expansion 100%
• Information Contraction 100%
• Information Graph 100%
• Prediction Task 50%
• Long-range Interactions 50%
• Computational Graph 50%
• Expansion Rate 50%
• Von Neumann 50%
• Signal Propagation 50%
• Graph Construction 50%
• Message Passing 50%
• Node Degree 50%
• Expansion Property 50%
• Signal Quality 50%
• Node Features 50%
• Spectral Expansion 50%
• Rewiring Strategy 50%
• Noisy Computation 50%
• Graph Rewiring 50%
• Reliable Computation 50%
• Edge Flip 50%
• Curvature-based 50%
• Synthetic Experiment 50%

Computer Science

• Graph Neural Network 100%
• Computation Graph 50%
• Signal Propagation 50%
• Message Passing 50%
• Graph Construction 50%
• Range Interaction 50%