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.
@article{arxiv.2206.02617,
title = {Individual Privacy Accounting for Differentially Private Stochastic Gradient Descent},
author = {Da Yu and Gautam Kamath and Janardhan Kulkarni and Tie-Yan Liu and Jian Yin and Huishuai Zhang},
journal= {arXiv preprint arXiv:2206.02617},
year = {2024}
}
Comments
Add clarification about the applicability of Definition 4