How to characterize the landscape of overparameterized convolutional neural networks

Keyphrases

• Feature Distribution 100%
• Convolutional Neural Network 100%
• Neural Network 40%
• Hidden Node 40%
• Hidden Layer 40%
• Continuous Limit 20%
• Process Feature 20%
• Nonconvex 20%
• Training Process 20%
• Convex Function 20%
• Initialization Method 20%
• Number of Channels 20%
• Network Parameters 20%
• Training Parameters 20%
• Deep Neural Network 20%
• Loss Landscape 20%
• Fixed Trajectory 20%

Computer Science

• Convolutional Neural Network 100%
• Neural Network 50%
• Training Process 25%
• Convex Function 25%
• Network Parameter 25%
• Deep Neural Network 25%

Neuroscience

• Neural Network 100%
• Grafting 16%