Algorithmic Stability of Heavy-Tailed Stochastic Gradient Descent on Least Squares

Keyphrases

• Stochastic Gradient Descent 100%
• Least Squares 100%
• Heavy Tails 75%
• Generalization Error 50%
• Uniform Stability 50%
• Heaviness 50%
• Neural Network 25%
• Tight 25%
• Stochastic Optimization 25%
• Network Testing 25%
• Stochastic Differential Equations 25%
• Statistical Regularities 25%
• Behavioral Properties 25%
• Regularity Assumption 25%
• Tail Behavior 25%
• Stability Bound 25%
• Generalization Behavior 25%
• Quadratic Optimization Problem 25%
• Heavy-tailedness 25%
• Squared Loss 25%
• Surrogate Loss 25%
• Generalization Properties 25%
• Euler Discretization 25%

Mathematics

• Stochastics 100%
• Least Square 100%
• Heavy Tail 60%
• Variance 20%
• Stochastic Differential Equation 20%
• Neural Network 20%
• Discretization 20%