WHAT HAPPENS AFTER SGD REACHES ZERO LOSS? -A MATHEMATICAL FRAMEWORK

Mathematics

• Stochastics 100%
• Manifold 50%
• Loss Function 33%
• Upper Bound 33%
• Covariance 33%
• Linear Models 16%
• Regularization 16%
• Minimax 16%
• Stochastic Differential Equation 16%
• Deep Learning 16%
• Like Term 16%
• Local Minimizer 16%

Keyphrases

• Track Gradient 16%

Computer Science

• Global Analysis 12%