Individual Privacy Accounting for Differentially Private Stochastic Gradient Descent

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yu, Da, Kamath, Gautam, Kulkarni, Janardhan, Liu, Tie-Yan, Yin, Jian, Zhang, Huishuai
Format: Preprint
Published: 2022
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916335408119808
author Yu, Da
Kamath, Gautam
Kulkarni, Janardhan
Liu, Tie-Yan
Yin, Jian
Zhang, Huishuai
author_facet Yu, Da
Kamath, Gautam
Kulkarni, Janardhan
Liu, Tie-Yan
Yin, Jian
Zhang, Huishuai
contents 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 $(\varepsilon,δ)$-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 $\varepsilon$ of the class with the lowest test accuracy is 44.2\% higher than that of the class with the highest accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_2206_02617
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle Individual Privacy Accounting for Differentially Private Stochastic Gradient Descent
Yu, Da
Kamath, Gautam
Kulkarni, Janardhan
Liu, Tie-Yan
Yin, Jian
Zhang, Huishuai
Machine Learning
Cryptography and Security
Data Structures and Algorithms
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 $(\varepsilon,δ)$-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 $\varepsilon$ of the class with the lowest test accuracy is 44.2\% higher than that of the class with the highest accuracy.
title Individual Privacy Accounting for Differentially Private Stochastic Gradient Descent
topic Machine Learning
Cryptography and Security
Data Structures and Algorithms
url https://arxiv.org/abs/2206.02617