The Dynamics of Gradient Descent for Overparametrized Neural Networks

Keyphrases

• Alternative Proof 20%
• Gradient Descent 100%
• Linear Approximation 20%
• Loss Function 20%
• Lyapunov Analysis 20%
• Minimum Norm Solution 20%
• Neural Network 20%
• Neural Network Weights 20%
• Overparametrized Neural Network 100%
• Parameter Values 20%
• Prediction Function 20%
• Single Hidden Layer Neural Network 20%
• Squared Loss 20%
• Training Error 40%
• Zero-training 20%

Mathematics

• Function Prediction 25%
• Initial Condition 25%
• Linear Approximation 25%
• Loss Function 25%
• Network Weight 25%
• Neural Network 100%