On Connecting Stochastic Gradient MCMC and Differential Privacy

Bai Li, Changyou Chen, Hao Liu, Lawrence Carin
Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, PMLR 89:557-566, 2019.

Abstract

Concerns related to data security and confidentiality have been raised when applying machine learning to real-world applications. Differential privacy provides a principled and rigorous privacy guarantee for machine learning models. While it is common to inject noise to design a model satisfying a required differential-privacy property, it is generally hard to balance the trade-off between privacy and utility. We show that stochastic gradient Markov chain Monte Carlo (SG-MCMC) – a class of scalable Bayesian posterior sampling algorithms – satisfies strong differential privacy, when carefully chosen stepsizes are employed. We develop theory on the performance of the proposed differentially-private SG-MCMC method. We conduct experiments to support our analysis, and show that a standard SG-MCMC sampler with minor modification can reach state-of-the-art performance in terms of both privacy and utility on Bayesian learning.

Cite this Paper


BibTeX
@InProceedings{pmlr-v89-li19a, title = {On Connecting Stochastic Gradient MCMC and Differential Privacy}, author = {Li, Bai and Chen, Changyou and Liu, Hao and Carin, Lawrence}, booktitle = {Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics}, pages = {557--566}, 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/li19a/li19a.pdf}, url = {https://proceedings.mlr.press/v89/li19a.html}, abstract = {Concerns related to data security and confidentiality have been raised when applying machine learning to real-world applications. Differential privacy provides a principled and rigorous privacy guarantee for machine learning models. While it is common to inject noise to design a model satisfying a required differential-privacy property, it is generally hard to balance the trade-off between privacy and utility. We show that stochastic gradient Markov chain Monte Carlo (SG-MCMC) – a class of scalable Bayesian posterior sampling algorithms – satisfies strong differential privacy, when carefully chosen stepsizes are employed. We develop theory on the performance of the proposed differentially-private SG-MCMC method. We conduct experiments to support our analysis, and show that a standard SG-MCMC sampler with minor modification can reach state-of-the-art performance in terms of both privacy and utility on Bayesian learning.} }
Endnote
%0 Conference Paper %T On Connecting Stochastic Gradient MCMC and Differential Privacy %A Bai Li %A Changyou Chen %A Hao Liu %A Lawrence Carin %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-li19a %I PMLR %P 557--566 %U https://proceedings.mlr.press/v89/li19a.html %V 89 %X Concerns related to data security and confidentiality have been raised when applying machine learning to real-world applications. Differential privacy provides a principled and rigorous privacy guarantee for machine learning models. While it is common to inject noise to design a model satisfying a required differential-privacy property, it is generally hard to balance the trade-off between privacy and utility. We show that stochastic gradient Markov chain Monte Carlo (SG-MCMC) – a class of scalable Bayesian posterior sampling algorithms – satisfies strong differential privacy, when carefully chosen stepsizes are employed. We develop theory on the performance of the proposed differentially-private SG-MCMC method. We conduct experiments to support our analysis, and show that a standard SG-MCMC sampler with minor modification can reach state-of-the-art performance in terms of both privacy and utility on Bayesian learning.
APA
Li, B., Chen, C., Liu, H. & Carin, L.. (2019). On Connecting Stochastic Gradient MCMC and Differential Privacy. Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 89:557-566 Available from https://proceedings.mlr.press/v89/li19a.html.

Related Material