Differentially private SGD with non-smooth losses

Mathematics

• Computational Complexity 8%
• Convex Optimization 9%
• Descent Algorithm 36%
• Excess 20%
• Generalization 4%
• Gradient 36%
• Gradient Algorithm 34%
• Gradient Descent 94%
• Least Squares 8%
• Lipschitz 7%
• Lipschitz Continuity 11%
• Logarithmic 7%
• Loss Function 9%
• Model 3%
• Output 7%
• Performance 5%
• Perturbation 13%
• Privacy 36%
• Robust Regression 12%
• Smoothness 31%
• Stochastic Gradient 100%
• Stochastic Optimization 11%
• Sufficient 6%
• Term 5%
• Trade-offs 9%
• Unbounded Domain 9%
• Violate 11%

Engineering & Materials Science

• Computational complexity 6%
• Convex optimization 8%
• Hinges 8%