Towards Deepening Graph Neural Networks: A GNTK-based Optimization Perspective

Keyphrases

• Kernel-based 100%
• Graph Neural Network 100%
• Neural Tangent Kernel 100%
• Trainability 100%
• Deep Graph Convolutional Network 60%
• Exponential Decay 40%
• Gradient Descent 40%
• Graph Convolutional Network 40%
• Graph Connectivity 20%
• Graph Structure 20%
• Asymptotic Behavior 20%
• Exponential Rate 20%
• Large Depth 20%
• Node Representation 20%
• Process Optimization 20%
• Expressive Power 20%
• Connection-based 20%
• Theoretical Research 20%
• Over-smoothing Problem 20%
• Connectivity-aware 20%
• Adaptive Sampling Strategy 20%
• Finite Width 20%
• Residual Connection 20%
• DropEdge 20%
• Trajectory Optimization 20%

Computer Science

• Graph Neural Network 100%
• Graph Convolutional Network 100%
• Gradient Descent 40%
• Exponential Decay 40%
• Theoretical Framework 20%
• Expressive Power 20%
• Fundamental Question 20%
• Structured Data 20%
• Process Optimization 20%
• Theoretical Research 20%

Chemical Engineering

• Neural Network 100%