A Differentially Private Method for Distributed Optimization in Directed Networks via State Decomposition

被引:23
作者
Chen, Xiaomeng [1 ]
Huang, Lingying [1 ,2 ]
He, Lidong [3 ]
Dey, Subhrakanti [4 ]
Shi, Ling [1 ]
机构
[1] Hong Kong Univ Sci & Technol, Dept Elect & Comp Engn, Hong Kong, Peoples R China
[2] Nanyang Technol Univ, Sch Elect & Elect Engn, Singapore 639798, Singapore
[3] Nanjing Univ Sci & Technol, Sch Automat, Nanjing, Peoples R China
[4] Uppsala Univ, Dept Elect Engn, S-75103 Uppsala, Sweden
来源
IEEE TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS | 2023年 / 10卷 / 04期
关键词
Privacy; Optimization; Differential privacy; Directed graphs; Convergence; Network systems; Distributed algorithms; Decomposition; differentially private; directed graph; distributed optimization; ECONOMIC-DISPATCH; AVERAGE CONSENSUS; ALGORITHM; CONVERGENCE;
D O I
10.1109/TCNS.2023.3264932
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
In this article, we study the problem of consensus-based distributed optimization, where a network of agents, abstracted as a directed graph, aims to minimize the sum of all agents' cost functions collaboratively. In the existing distributed optimization approaches (Push-Pull/AB) for directed graphs, all the agents exchange their states with neighbors to achieve the optimal solution with a constant step size, which may lead to the disclosure of sensitive and private information. For privacy preservation, we propose a novel state-decomposition-based gradient tracking approach (SD-Push-Pull) for distributed optimization over directed networks that preserves differential privacy, which is a strong notion that protects agents' privacy against an adversary with arbitrary auxiliary information. The main idea of the proposed approach is to decompose the gradient state of each agent into two substates. Only one substate is exchanged by the agent with its neighbors over time, and the other one is not shared. That is to say, only one substate is visible to an adversary, protecting the sensitive information from being leaked. It is proved that under certain decomposition principles, a bound for the suboptimality of the proposed algorithm can be derived, and the differential privacy is achieved simultaneously. Moreover, the tradeoff between differential privacy and the optimization accuracy is also characterized. Finally, a numerical simulation is provided to illustrate the effectiveness of the proposed approach.
引用
收藏
页码:2165 / 2177
页数:13
相关论文
共 30 条
[11]   Privacy preserving distributed optimization using homomorphic encryption [J].
Lu, Yang ;
Zhu, Minghui .
AUTOMATICA, 2018, 96 :314-325
[12]   Privacy Preserving Consensus-Based Economic Dispatch in Smart Grid Systems [J].
Mandal, Avikarsha .
FUTURE NETWORK SYSTEMS AND SECURITY, 2016, 670 :98-110
[13]   A Privacy Preserving Distributed Optimization Algorithm for Economic Dispatch Over Time-Varying Directed Networks [J].
Mao, Shuai ;
Tang, Yang ;
Dong, Ziwei ;
Meng, Ke ;
Dong, Zhao Yang ;
Qian, Feng .
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, 2021, 17 (03) :1689-1701
[14]   Distributed Optimization and Coordination Algorithms for Dynamic Traffic Metering in Urban Street Networks [J].
Mohebifard, Rasool ;
Hajbabaie, Ali .
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, 2019, 20 (05) :1930-1941
[15]   ACHIEVING GEOMETRIC CONVERGENCE FOR DISTRIBUTED OPTIMIZATION OVER TIME-VARYING GRAPHS [J].
Nedic, Angelia ;
Olshevsky, Alex ;
Shi, Wei .
SIAM JOURNAL ON OPTIMIZATION, 2017, 27 (04) :2597-2633
[16]   Distributed Subgradient Methods for Multi-Agent Optimization [J].
Nedic, Angelia ;
Ozdaglar, Asurrian .
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, 2009, 54 (01) :48-61
[17]   Differentially private average consensus: Obstructions, trade-offs, and optimal algorithm design [J].
Nozari, Erfan ;
Tallapragada, Pavankumar ;
Cortes, Jorge .
AUTOMATICA, 2017, 81 :221-231
[18]  
Polyak B.T.:., 1987, Introduction to optimization, V1, P32
[19]  
Pu S, 2020, IEEE DECIS CONTR P, P2335, DOI [10.1109/CDC42340.2020.9303917, 10.1109/cdc42340.2020.9303917]
[20]   Distributed stochastic gradient tracking methods [J].
Pu, Shi ;
Nedic, Angelia .
MATHEMATICAL PROGRAMMING, 2021, 187 (1-2) :409-457