On the cost of differential privacy in distributed control systems

Zhenqi Huang, Yu Wang, Sayan Mitra, Geir E Dullerud

Research output: Contribution to conferencePaper

Abstract

Individuals sharing information can improve the cost or performance of a distributed control system. But, sharing may also violate privacy. We develop a general framework for studying the cost of differential privacy in systems where a collection of agents, with coupled dynamics, communicate for sensing their shared environment while pursuing individual preferences. First, we propose a communication strategy that relies on adding carefully chosen random noise to agent states and show that it preserves differential privacy. Of course, the higher the standard deviation of the noise, the higher the cost of privacy. For linear distributed control systems with quadratic cost functions, the standard deviation becomes independent of the number agents and it decays with the maximum eigenvalue of the dynamics matrix. Furthermore, for stable dynamics, the noise to be added is independent of the number of agents as well as the time horizon up to which privacy is desired. Finally, we show that the cost of ε-differential privacy up to time T, for a linear stable system with N agents, is upper bounded by O(T3/Nε2 ).

Original languageEnglish (US)
Pages105-114
Number of pages10
DOIs
StatePublished - Jan 1 2014
Event2014 3rd ACM International Conference on High Confidence Networked Systems, HiCoNS 2014, Part of CPSWeek 2014 - Berlin, Germany
Duration: Apr 15 2014Apr 17 2014

Other

Other2014 3rd ACM International Conference on High Confidence Networked Systems, HiCoNS 2014, Part of CPSWeek 2014
CountryGermany
CityBerlin
Period4/15/144/17/14

Fingerprint

Distributed parameter control systems
Costs
Linear control systems
Cost functions
Communication

Keywords

  • Cyber-physical security
  • Differential privacy
  • Distributed control

ASJC Scopus subject areas

  • Computer Networks and Communications

Cite this

Huang, Z., Wang, Y., Mitra, S., & Dullerud, G. E. (2014). On the cost of differential privacy in distributed control systems. 105-114. Paper presented at 2014 3rd ACM International Conference on High Confidence Networked Systems, HiCoNS 2014, Part of CPSWeek 2014, Berlin, Germany. https://doi.org/10.1145/2566468.2566474

On the cost of differential privacy in distributed control systems. / Huang, Zhenqi; Wang, Yu; Mitra, Sayan; Dullerud, Geir E.

2014. 105-114 Paper presented at 2014 3rd ACM International Conference on High Confidence Networked Systems, HiCoNS 2014, Part of CPSWeek 2014, Berlin, Germany.

Research output: Contribution to conferencePaper

Huang, Z, Wang, Y, Mitra, S & Dullerud, GE 2014, 'On the cost of differential privacy in distributed control systems' Paper presented at 2014 3rd ACM International Conference on High Confidence Networked Systems, HiCoNS 2014, Part of CPSWeek 2014, Berlin, Germany, 4/15/14 - 4/17/14, pp. 105-114. https://doi.org/10.1145/2566468.2566474
Huang Z, Wang Y, Mitra S, Dullerud GE. On the cost of differential privacy in distributed control systems. 2014. Paper presented at 2014 3rd ACM International Conference on High Confidence Networked Systems, HiCoNS 2014, Part of CPSWeek 2014, Berlin, Germany. https://doi.org/10.1145/2566468.2566474
Huang, Zhenqi ; Wang, Yu ; Mitra, Sayan ; Dullerud, Geir E. / On the cost of differential privacy in distributed control systems. Paper presented at 2014 3rd ACM International Conference on High Confidence Networked Systems, HiCoNS 2014, Part of CPSWeek 2014, Berlin, Germany.10 p.
@conference{f9fb92161d7d423396ea9cc071c52427,
title = "On the cost of differential privacy in distributed control systems",
abstract = "Individuals sharing information can improve the cost or performance of a distributed control system. But, sharing may also violate privacy. We develop a general framework for studying the cost of differential privacy in systems where a collection of agents, with coupled dynamics, communicate for sensing their shared environment while pursuing individual preferences. First, we propose a communication strategy that relies on adding carefully chosen random noise to agent states and show that it preserves differential privacy. Of course, the higher the standard deviation of the noise, the higher the cost of privacy. For linear distributed control systems with quadratic cost functions, the standard deviation becomes independent of the number agents and it decays with the maximum eigenvalue of the dynamics matrix. Furthermore, for stable dynamics, the noise to be added is independent of the number of agents as well as the time horizon up to which privacy is desired. Finally, we show that the cost of ε-differential privacy up to time T, for a linear stable system with N agents, is upper bounded by O(T3/Nε2 ).",
keywords = "Cyber-physical security, Differential privacy, Distributed control",
author = "Zhenqi Huang and Yu Wang and Sayan Mitra and Dullerud, {Geir E}",
year = "2014",
month = "1",
day = "1",
doi = "10.1145/2566468.2566474",
language = "English (US)",
pages = "105--114",
note = "2014 3rd ACM International Conference on High Confidence Networked Systems, HiCoNS 2014, Part of CPSWeek 2014 ; Conference date: 15-04-2014 Through 17-04-2014",

}

TY - CONF

T1 - On the cost of differential privacy in distributed control systems

AU - Huang, Zhenqi

AU - Wang, Yu

AU - Mitra, Sayan

AU - Dullerud, Geir E

PY - 2014/1/1

Y1 - 2014/1/1

N2 - Individuals sharing information can improve the cost or performance of a distributed control system. But, sharing may also violate privacy. We develop a general framework for studying the cost of differential privacy in systems where a collection of agents, with coupled dynamics, communicate for sensing their shared environment while pursuing individual preferences. First, we propose a communication strategy that relies on adding carefully chosen random noise to agent states and show that it preserves differential privacy. Of course, the higher the standard deviation of the noise, the higher the cost of privacy. For linear distributed control systems with quadratic cost functions, the standard deviation becomes independent of the number agents and it decays with the maximum eigenvalue of the dynamics matrix. Furthermore, for stable dynamics, the noise to be added is independent of the number of agents as well as the time horizon up to which privacy is desired. Finally, we show that the cost of ε-differential privacy up to time T, for a linear stable system with N agents, is upper bounded by O(T3/Nε2 ).

AB - Individuals sharing information can improve the cost or performance of a distributed control system. But, sharing may also violate privacy. We develop a general framework for studying the cost of differential privacy in systems where a collection of agents, with coupled dynamics, communicate for sensing their shared environment while pursuing individual preferences. First, we propose a communication strategy that relies on adding carefully chosen random noise to agent states and show that it preserves differential privacy. Of course, the higher the standard deviation of the noise, the higher the cost of privacy. For linear distributed control systems with quadratic cost functions, the standard deviation becomes independent of the number agents and it decays with the maximum eigenvalue of the dynamics matrix. Furthermore, for stable dynamics, the noise to be added is independent of the number of agents as well as the time horizon up to which privacy is desired. Finally, we show that the cost of ε-differential privacy up to time T, for a linear stable system with N agents, is upper bounded by O(T3/Nε2 ).

KW - Cyber-physical security

KW - Differential privacy

KW - Distributed control

UR - http://www.scopus.com/inward/record.url?scp=84899756844&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84899756844&partnerID=8YFLogxK

U2 - 10.1145/2566468.2566474

DO - 10.1145/2566468.2566474

M3 - Paper

AN - SCOPUS:84899756844

SP - 105

EP - 114

ER -