Differentially private SGD with non-smooth losses

Keyphrases

• Nonsmooth 100%
• Stochastic Gradient Algorithm 100%
• Smooth Loss 100%
• Differentially Private Stochastic Gradient Descent 100%
• Stochastic Gradient Descent 66%
• Differential Privacy 66%
• Lipschitz Continuity 66%
• Gradient Complexity 66%
• Gradient Perturbation 66%
• Computational Complexity 33%
• Popular 33%
• Unbounded Domain 33%
• Uniformly Bounded 33%
• Linear Gradient 33%
• Optimal Performance 33%
• Loss Function 33%
• Logarithmic Term 33%
• Utility Guarantees 33%
• Hinge Loss 33%
• Robust Regression 33%
• Non-convex Cost Function 33%
• Population Risk 33%
• Excess Risk 33%
• Risk Formula 33%
• Stochastic Convex Optimization 33%
• Continuous Gradient 33%
• Least Squares Loss 33%
• Output Perturbation 33%

Mathematics

• Stochastics 100%
• Unbounded Domain 14%
• Loss Function 14%
• Logarithmic Term 14%
• Excess Risk 14%
• Robust Regression 14%
• Lipschitz Continuity 14%
• Square Loss 14%
• Least Squares Method 14%