DPSUR: Accelerating Differentially Private Stochastic Gradient Descent Using Selective Update and Release

Keyphrases

• Selective Release 100%
• Selective Update 100%
• Differentially Private Stochastic Gradient Descent 100%
• Gradient Selection 66%
• Privacy Concerns 33%
• De Facto Standard 33%
• Selection Strategy 33%
• Test Validity 33%
• Fast Convergence 33%
• Gaussian Noise 33%
• Training Algorithm 33%
• Slow Convergence 33%
• Training Model 33%
• Privacy Attack 33%
• Machine Learning Models 33%
• Differential Privacy 33%
• Model Update 33%
• Training Framework 33%
• Differentially Private 33%
• Stochastic Gradient Descent 33%
• Utility Loss 33%
• Private Data 33%
• Training Loss 33%
• Membership Inference 33%
• Privacy-preserving Machine Learning 33%
• Model Inversion 33%
• Gradient Update 33%
• Private Training 33%
• Gradient Evaluation 33%
• Popular 33%
• Threshold Mechanism 33%

Computer Science

• Gradient Descent 100%
• De Facto Standard 50%
• Training Model 50%
• Differential Privacy 50%
• Privacy Concern 50%
• Slow Convergence 50%
• Fast Convergence 50%
• training algorithm 50%
• Privacy Preserving Machine Learning 50%
• Selection Strategy 50%
• Machine Learning 50%
• Learning System 50%
• Gaussian White Noise 50%