Abstract

Many applications of machine learning, for example in health care, would benefit from methods that can guarantee privacy of data subjects. Differential privacy (DP) has become established as a standard for protecting learning results. The standard DP algorithms require a single trusted party to have access to the entire data, which is a clear weakness, or add prohibitive amounts of noise. We consider DP Bayesian learning in a distributed setting, where each party only holds a single sample or a few samples of the data. We propose a learning strategy based on a secure multi-party sum function for aggregating summaries from data holders and the Gaussian mechanism for DP. Our method builds on an asymptotically optimal and practically efficient DP Bayesian inference with rapidly diminishing extra cost.
Original languageEnglish
Title of host publicationAdvances in Neural Information Processing Systems 30 (NIPS 2017)
EditorsI. Guyon, U.V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, R. Garnett
Number of pages10
Volume30
PublisherNEURAL INFORMATION PROCESSING SYSTEMS (NIPS)
Publication date2017
Publication statusPublished - 2017
MoE publication typeA4 Article in conference proceedings
EventAnnual Conference on Neural Information Processing Systems - Long Beach, United States
Duration: 4 Dec 20179 Dec 2017
Conference number: 31
http://nips.cc/Conferences/2017

Publication series

NameAdvances in Neural Information Processing Systems
PublisherNEURAL INFORMATION PROCESSING SYSTEMS (NIPS)
Volume30
ISSN (Print)1049-5258

Fields of Science

  • NOISE
  • 112 Statistics and probability
  • 113 Computer and information sciences

Cite this

Heikkila, M., Lagerspetz, E., Kaski, S., Shimizu, K., Tarkoma, S., & Honkela, A. (2017). Differentially private Bayesian learning on distributed data. In I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, & R. Garnett (Eds.), Advances in Neural Information Processing Systems 30 (NIPS 2017) (Vol. 30). (Advances in Neural Information Processing Systems; Vol. 30). NEURAL INFORMATION PROCESSING SYSTEMS (NIPS).
Heikkila, Mikko ; Lagerspetz, Eemil ; Kaski, Samuel ; Shimizu, Kana ; Tarkoma, Sasu ; Honkela, Antti. / Differentially private Bayesian learning on distributed data. Advances in Neural Information Processing Systems 30 (NIPS 2017). editor / I. Guyon ; U.V. Luxburg ; S. Bengio ; H. Wallach ; R. Fergus ; S. Vishwanathan ; R. Garnett. Vol. 30 NEURAL INFORMATION PROCESSING SYSTEMS (NIPS), 2017. (Advances in Neural Information Processing Systems).
@inproceedings{854e468f4c0c4f6c9f0bcb1f3bd252d6,
title = "Differentially private Bayesian learning on distributed data",
abstract = "Many applications of machine learning, for example in health care, would benefit from methods that can guarantee privacy of data subjects. Differential privacy (DP) has become established as a standard for protecting learning results. The standard DP algorithms require a single trusted party to have access to the entire data, which is a clear weakness, or add prohibitive amounts of noise. We consider DP Bayesian learning in a distributed setting, where each party only holds a single sample or a few samples of the data. We propose a learning strategy based on a secure multi-party sum function for aggregating summaries from data holders and the Gaussian mechanism for DP. Our method builds on an asymptotically optimal and practically efficient DP Bayesian inference with rapidly diminishing extra cost.",
keywords = "NOISE, 112 Statistics and probability, 113 Computer and information sciences",
author = "Mikko Heikkila and Eemil Lagerspetz and Samuel Kaski and Kana Shimizu and Sasu Tarkoma and Antti Honkela",
year = "2017",
language = "English",
volume = "30",
series = "Advances in Neural Information Processing Systems",
publisher = "NEURAL INFORMATION PROCESSING SYSTEMS (NIPS)",
editor = "I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett",
booktitle = "Advances in Neural Information Processing Systems 30 (NIPS 2017)",
address = "United States",

}

Heikkila, M, Lagerspetz, E, Kaski, S, Shimizu, K, Tarkoma, S & Honkela, A 2017, Differentially private Bayesian learning on distributed data. in I Guyon, UV Luxburg, S Bengio, H Wallach, R Fergus, S Vishwanathan & R Garnett (eds), Advances in Neural Information Processing Systems 30 (NIPS 2017). vol. 30, Advances in Neural Information Processing Systems, vol. 30, NEURAL INFORMATION PROCESSING SYSTEMS (NIPS), Annual Conference on Neural Information Processing Systems, Long Beach, United States, 04/12/2017.

Differentially private Bayesian learning on distributed data. / Heikkila, Mikko; Lagerspetz, Eemil; Kaski, Samuel; Shimizu, Kana; Tarkoma, Sasu; Honkela, Antti.

Advances in Neural Information Processing Systems 30 (NIPS 2017). ed. / I. Guyon; U.V. Luxburg; S. Bengio; H. Wallach; R. Fergus; S. Vishwanathan; R. Garnett. Vol. 30 NEURAL INFORMATION PROCESSING SYSTEMS (NIPS), 2017. (Advances in Neural Information Processing Systems; Vol. 30).

Research output: Chapter in Book/Report/Conference proceedingConference contributionScientificpeer-review

TY - GEN

T1 - Differentially private Bayesian learning on distributed data

AU - Heikkila, Mikko

AU - Lagerspetz, Eemil

AU - Kaski, Samuel

AU - Shimizu, Kana

AU - Tarkoma, Sasu

AU - Honkela, Antti

PY - 2017

Y1 - 2017

N2 - Many applications of machine learning, for example in health care, would benefit from methods that can guarantee privacy of data subjects. Differential privacy (DP) has become established as a standard for protecting learning results. The standard DP algorithms require a single trusted party to have access to the entire data, which is a clear weakness, or add prohibitive amounts of noise. We consider DP Bayesian learning in a distributed setting, where each party only holds a single sample or a few samples of the data. We propose a learning strategy based on a secure multi-party sum function for aggregating summaries from data holders and the Gaussian mechanism for DP. Our method builds on an asymptotically optimal and practically efficient DP Bayesian inference with rapidly diminishing extra cost.

AB - Many applications of machine learning, for example in health care, would benefit from methods that can guarantee privacy of data subjects. Differential privacy (DP) has become established as a standard for protecting learning results. The standard DP algorithms require a single trusted party to have access to the entire data, which is a clear weakness, or add prohibitive amounts of noise. We consider DP Bayesian learning in a distributed setting, where each party only holds a single sample or a few samples of the data. We propose a learning strategy based on a secure multi-party sum function for aggregating summaries from data holders and the Gaussian mechanism for DP. Our method builds on an asymptotically optimal and practically efficient DP Bayesian inference with rapidly diminishing extra cost.

KW - NOISE

KW - 112 Statistics and probability

KW - 113 Computer and information sciences

M3 - Conference contribution

VL - 30

T3 - Advances in Neural Information Processing Systems

BT - Advances in Neural Information Processing Systems 30 (NIPS 2017)

A2 - Guyon, I.

A2 - Luxburg, U.V.

A2 - Bengio, S.

A2 - Wallach, H.

A2 - Fergus, R.

A2 - Vishwanathan, S.

A2 - Garnett, R.

PB - NEURAL INFORMATION PROCESSING SYSTEMS (NIPS)

ER -

Heikkila M, Lagerspetz E, Kaski S, Shimizu K, Tarkoma S, Honkela A. Differentially private Bayesian learning on distributed data. In Guyon I, Luxburg UV, Bengio S, Wallach H, Fergus R, Vishwanathan S, Garnett R, editors, Advances in Neural Information Processing Systems 30 (NIPS 2017). Vol. 30. NEURAL INFORMATION PROCESSING SYSTEMS (NIPS). 2017. (Advances in Neural Information Processing Systems).