Predicting the outputs of finite deep neural networks trained with noisy gradients

Keyphrases

• Deep Neural Network 100%
• Gaussian Process 83%
• Finite Width 33%
• Neural Tangent Kernel 33%
• Finite Width Correction 33%
• Neural Network Gaussian Processes 33%
• Analytical Framework 16%
• Training Protocol 16%
• Image Classification 16%
• Activation Function 16%
• Analytical Form 16%
• Classification Task 16%
• Convolutional Neural Network 16%
• Training Sample Size 16%
• Decay Width 16%
• Weight Decay 16%
• Finite Networks 16%
• Weight Map 16%
• Gradient Flow 16%
• Deep Neural Network Training 16%
• Infinite-width Limit 16%
• Non-Gaussian Stochastic Processes 16%
• Non-Gaussian Process 16%
• Process Error 16%

Mathematics

• Deep Neural Network 100%
• Gaussian Process 100%
• Neural Network 25%
• Set Size 12%
• Training Set 12%
• Gradient Flow 12%
• Gaussian Stochastic Process 12%
• Process Error 12%
• Convolutional Neural Network 12%