Distributed Stochastic Optimization via Matrix Exponential Learning

被引:28
作者
Mertikopoulos, Panayotis [1 ]
Belmega, E. Veronica [2 ,3 ]
Negrel, Romain [4 ,5 ]
Sanguinetti, Luca [6 ,7 ]
机构
[1] French Natl Ctr Sci Res, Lab Informat Grenoble, F-38000 Grenoble, France
[2] ETIS ENSEA Univ Cergy Pontoise CNRS, F-95014 Cergy Pontoise, France
[3] Inria, F-38334 Grenoble, France
[4] Univ Caen Normandie, UNSCICAEN, GREYC UMR 6072, F-14032 Caen, France
[5] Univ Paris Est, LIGM, Equipe A3SI, ESIEE, F-77454 Marne La Vallee, France
[6] Univ Pisa, Dipartimento Ingn Informaz, I-56100 Pisa, Italy
[7] Univ Paris Saclay, Large Syst & Networks Grp, CentraleSupelec, F-91190 Gif Sur Yvette, France
基金
欧洲研究理事会;
关键词
Learning; stochastic optimization; game theory; matrix exponential learning; variational stability; uncertainty; IMAGE RETRIEVAL; MIMO; GAME; DYNAMICS;
D O I
10.1109/TSP.2017.2656847
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 this paper, we investigate a distributed learning scheme for a broad class of stochastic optimization problems and games that arise in signal processing and wireless communications. The proposed algorithm relies on the method of matrix exponential learning (MXL) and only requires locally computable gradient observations that are possibly imperfect. To analyze it, we introduce the notion of a stable Nash equilibrium and we show that the algorithm is globally convergent to such equilibria-or locally convergent when an equilibrium is only locally stable. To complement our convergence analysis, we also derive explicit bounds for the algorithm's convergence speed and we test it in realistic multicarrier/multiple-antenna wireless scenarios where several users seek to maximize their energy efficiency. Our results show that learning allows users to attain a net increase between 100% and 500% in energy efficiency, even under very high uncertainty.
引用
收藏
页码:2277 / 2290
页数:14
相关论文
共 42 条
[1]  
[Anonymous], 2016, LEARNING CONCAVE GAM
[2]  
[Anonymous], 2014, TECH REP
[3]  
[Anonymous], 2003, SPRINGER SERIES OPER
[4]  
[Anonymous], CISC VIS NETW IND GL
[5]  
[Anonymous], P 30 INT C MACH LEAR
[6]  
[Anonymous], P IEEE GLOB C SIGN I
[7]  
[Anonymous], DYNAMIC POWER ALLOCA
[8]  
[Anonymous], 2004, INTRO LECT CONVEX OP
[9]  
[Anonymous], WORKING PAPER
[10]  
[Anonymous], 2009, Stochastic Approximation: A Dynamical Systems Viewpoint