Individual Privacy Accounting for Differentially Private Stochastic Gradient Descent

2022/06/06 by Yu, Da, Kamath, Gautam, Kulkarni, Janardhan +3
#Cryptography and Security (cs.CR) #Data Structures and Algorithms (cs.DS) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML)

paper · doi:10.48550/arxiv.2206.02617

Abstract

Differentially private stochastic gradient descent (DP-SGD) is the workhorse algorithm for recent advances in private deep learning. It provides a single privacy guarantee to all datapoints in the dataset. We propose output-specific (ε,δ)-DP to characterize privacy guarantees for individual examples when releasing models trained by DP-SGD. We also design an efficient algorithm to investigate individual privacy across a number of datasets. We find that most examples enjoy stronger privacy guarantees than the worst-case bound. We further discover that the training loss and the privacy parameter of an example are well-correlated. This implies groups that are underserved in terms of model utility simultaneously experience weaker privacy guarantees. For example, on CIFAR-10, the average ε of the class with the lowest test accuracy is 44.2% higher than that of the class with the highest accuracy.

Related

• Gradients Look Alike: Sensitivity is Often Overestimated in DP-SGD
• Have it your way: Individualized Privacy Assignment for DP-SGD
• Personalized DP-SGD using Sampling Mechanisms
• Too Good to be True? Turn Any Model Differentially Private With DP-Weights
• An Improved Privacy and Utility Analysis of Differentially Private SGD with Bounded Domain and Smooth Losses