Subsampled Renyi Differential Privacy and Analytical Moments Accountant

Yu-Xiang Wang, Borja Balle, Shiva Prasad Kasiviswanathan
Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, PMLR 89:1226-1235, 2019.

Abstract

We study the problem of subsampling in differential privacy (DP), a question that is the centerpiece behind many successful differentially private machine learning algorithms. Specifically, we provide a tight upper bound on the Renyi Differential Privacy (RDP) [Mironov 2017] parameters for algorithms that: (1) subsample the dataset, and then (2) applies a randomized mechanism M to the subsample, in terms of the RDP parameters of M and the subsampling probability parameter. Our results generalize the moments accounting technique, developed by [Abadi et al. 2016] for the Gaussian mechanism, to any subsampled RDP mechanism.

Cite this Paper


BibTeX
@InProceedings{pmlr-v89-wang19b, title = {Subsampled Renyi Differential Privacy and Analytical Moments Accountant}, author = {Wang, Yu-Xiang and Balle, Borja and Kasiviswanathan, Shiva Prasad}, booktitle = {Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics}, pages = {1226--1235}, year = {2019}, editor = {Chaudhuri, Kamalika and Sugiyama, Masashi}, volume = {89}, series = {Proceedings of Machine Learning Research}, month = {16--18 Apr}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v89/wang19b/wang19b.pdf}, url = {https://proceedings.mlr.press/v89/wang19b.html}, abstract = {We study the problem of subsampling in differential privacy (DP), a question that is the centerpiece behind many successful differentially private machine learning algorithms. Specifically, we provide a tight upper bound on the Renyi Differential Privacy (RDP) [Mironov 2017] parameters for algorithms that: (1) subsample the dataset, and then (2) applies a randomized mechanism M to the subsample, in terms of the RDP parameters of M and the subsampling probability parameter. Our results generalize the moments accounting technique, developed by [Abadi et al. 2016] for the Gaussian mechanism, to any subsampled RDP mechanism.} }
Endnote
%0 Conference Paper %T Subsampled Renyi Differential Privacy and Analytical Moments Accountant %A Yu-Xiang Wang %A Borja Balle %A Shiva Prasad Kasiviswanathan %B Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2019 %E Kamalika Chaudhuri %E Masashi Sugiyama %F pmlr-v89-wang19b %I PMLR %P 1226--1235 %U https://proceedings.mlr.press/v89/wang19b.html %V 89 %X We study the problem of subsampling in differential privacy (DP), a question that is the centerpiece behind many successful differentially private machine learning algorithms. Specifically, we provide a tight upper bound on the Renyi Differential Privacy (RDP) [Mironov 2017] parameters for algorithms that: (1) subsample the dataset, and then (2) applies a randomized mechanism M to the subsample, in terms of the RDP parameters of M and the subsampling probability parameter. Our results generalize the moments accounting technique, developed by [Abadi et al. 2016] for the Gaussian mechanism, to any subsampled RDP mechanism.
APA
Wang, Y., Balle, B. & Kasiviswanathan, S.P.. (2019). Subsampled Renyi Differential Privacy and Analytical Moments Accountant. Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 89:1226-1235 Available from https://proceedings.mlr.press/v89/wang19b.html.

Related Material