Individual Privacy Accounting for Differentially Private Stochastic Gradient Descent
Yu, Da, Kamath, Gautam, Kulkarni, Janardhan, Liu, Tie-Yan, Yin, Jian, Zhang, Huishuai
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 $(\varepsilon,\delta)$-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 $\varepsilon$ of the class with the lowest test accuracy is 44.2\% higher than that of the class with the highest accuracy.
Sep-2-2023
- Genre:
- Research Report > New Finding (0.46)
- Industry:
- Information Technology > Security & Privacy (1.00)
- Technology: