Network Topology and Communication-Computation Tradeoffs in Decentralized Optimization

被引:376
作者
Nedic, Angelia [1 ]
Olshevsky, Alex [2 ]
Rabbat, Michael G. [3 ,4 ]
机构
[1] Arizona State Univ, Sch Elect Comp & Energy Engn, Tempe, AZ 85281 USA
[2] Boston Univ, Dept Elect & Comp Engn, Boston, MA 02215 USA
[3] Facebook Artificial Intelligence Res, Montreal, PQ, Canada
[4] McGill Univ, Dept Elect & Comp Engn, Montreal, PQ H3A 0G4, Canada
基金
美国国家科学基金会; 加拿大自然科学与工程研究理事会;
关键词
Consensus algorithms; distributed optimization; distributed averaging; gossip algorithms; multiagent systems; DISTRIBUTED CONSTRAINED OPTIMIZATION; MOBILE AUTONOMOUS AGENTS; AD-HOC WSNS; AVERAGE CONSENSUS; PROJECTION ALGORITHMS; GOSSIP ALGORITHMS; NOISY LINKS; TIME; STRATEGIES; CONVERGENCE;
D O I
10.1109/JPROC.2018.2817461
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
In decentralized optimization, nodes cooperate to minimize an overall objective function that is the sum (or average) of per-node private objective functions. Algorithms interleave local computations with communication among all or a subset of the nodes. Motivated by a variety of applications decentralized estimation in sensor networks, fitting models to massive data sets, and decentralized control of multirobot systems, to name a few significant advances have been made toward the development of robust, practical algorithms with theoretical performance guarantees. This paper presents an overview of recent work in this area. In general, rates of convergence depend not only on the number of nodes involved and the desired level of accuracy, but also on the structure and nature of the network over which nodes communicate (e.g., whether links are directed or undirected, static or time varying). We survey the state-of-the-art algorithms and their analyses tailored to these different scenarios, highlighting the role of the network topology.
引用
收藏
页码:953 / 976
页数:24
相关论文
共 116 条
[61]   Distributed multi-agent optimization with state-dependent communication [J].
Lobel, Ilan ;
Ozdaglar, Asuman ;
Feijer, Diego .
MATHEMATICAL PROGRAMMING, 2011, 129 (02) :255-284
[62]   Distributed Subgradient Methods for Convex Optimization Over Random Networks [J].
Lobel, Ilan ;
Ozdaglar, Asuman .
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, 2011, 56 (06) :1291-1306
[63]   Zero-Gradient-Sum Algorithms for Distributed Convex Optimization: The Continuous-Time Case [J].
Lu, Jie ;
Tang, Choon Yik .
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, 2012, 57 (09) :2348-2354
[64]   NONLINEAR GOSSIP [J].
Mathkar, Adwaitvedant S. ;
Borkar, Vivek S. .
SIAM JOURNAL ON CONTROL AND OPTIMIZATION, 2016, 54 (03) :1535-1557
[65]  
Nedic A, 2016, ACHIEVING GEOMETRIC
[66]   Stochastic Gradient-Push for Strongly Convex Functions on Time-Varying Directed Graphs [J].
Nedic, Angelia ;
Olshevsky, Alex .
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, 2016, 61 (12) :3936-3947
[67]  
Nedic A, 2015, P AMER CONTR CONF, P4497, DOI 10.1109/ACC.2015.7172037
[68]   Distributed Optimization Over Time-Varying Directed Graphs [J].
Nedic, Angelia ;
Olshevsky, Alex .
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, 2015, 60 (03) :601-615
[69]   Random algorithms for convex minimization problems [J].
Nedic, Angelia .
MATHEMATICAL PROGRAMMING, 2011, 129 (02) :225-253
[70]   Constrained Consensus and Optimization in Multi-Agent Networks [J].
Nedic, Angelia ;
Ozdaglar, Asuman ;
Parrilo, Pablo A. .
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, 2010, 55 (04) :922-938