Gradient descent finds global minima of deep neural networks

Keyphrases

• Global Minimum 100%
• Gradient Descent 100%
• Deep Neural Network 100%
• Gram Matrix 66%
• Polynomial Time 33%
• Global Optimality 33%
• Gradient Method 33%
• Non-convexity 33%
• Neural Network Architecture 33%
• Training Loss 33%
• Convergence Results 33%
• Training Process 33%
• Training Deep Neural Networks 33%
• Zero-training 33%
• Deep Residual Convolutional Neural Network 33%
• Overparametrized Neural Networks 33%
• Residual Connection 33%

Computer Science

• Gradient Descent 100%
• Deep Neural Network 100%
• Neural Network 25%
• Polynomial Time 25%
• Global Optimality 25%
• Objective Function 25%
• Neural Network Architecture 25%
• Training Process 25%
• Convolutional Neural Network 25%