Spectral-DP: Differentially Private Deep Learning through Spectral Perturbation and Filtering

Keyphrases

• Deep Learning 100%
• Spectral Filtering 100%
• Differentially Private 100%
• Spectral Perturbation 100%
• Differentially Private Stochastic Gradient Descent 100%
• Fully Connected Layer 66%
• Transfer Learning 33%
• Benchmark Dataset 33%
• Differential Privacy 33%
• Learning Environment 33%
• Spectral Domain 33%
• Learning Approaches 33%
• Comprehensive Experiment 33%
• Training Approaches 33%
• Noisy Training 33%
• Utility Performance 33%
• Deep Learning Algorithm 33%
• Dense Neural Network 33%
• Low Noise 33%
• Utility Costs 33%
• Noise Addition 33%
• Training from Scratch 33%
• Gradient Perturbation 33%
• Spatial Restructuring 33%
• Differentially Private Learning 33%
• Block Circulant 33%
• Privacy Guarantee 33%
• Direct Noise 33%

Computer Science

• Deep Learning Method 100%
• Stochastic Gradient Descent 75%
• Fully Connected Layer 50%
• Differential Privacy 25%
• Transfer Learning 25%
• Neural Network 25%
• Spectral Domain 25%
• Learning Approach 25%
• Noise Addition 25%
• Convolution 25%