ON THE DIVERGENCE OF DECENTRALIZED NONCONVEX OPTIMIZATION

被引:4
|
作者
Hong, M. I. N. G. Y. I. [1 ]
Zeng, S. I. L. I. A. N. G. [1 ]
Zhang, J. U. N. Y. U. [2 ]
Sun, H. A. O. R. A. N. [1 ]
机构
[1] Univ Minnesota Twin Cities, Dept ECE, Minneapolis, MN 55455 USA
[2] Natl Univ Singapore, Dept ISEM, Singapore 119007, Singapore
关键词
decentralized optimization; nonconvex problems; Lipschitz continuous gradient; DISTRIBUTED OPTIMIZATION; CONVERGENCE; ALGORITHM; STRATEGIES;
D O I
10.1137/20M1353149
中图分类号
O29 [应用数学];
学科分类号
070104 ;
摘要
agents jointly optimize the nonconvex objective function f(u) := 1/N\sumN Abstract. In this work, we study a generic class of decentralized algorithms in which N i=1 fi(u), while only communicating with their neighbors. This class of problems has become popular in modeling many signal processing and decentralized machine learning applications, and efficient algorithms have been proposed for such a type of problem. However, most of the existing decentralized algorithms require that the local function gradients V fi's as well as the average function gradient Vf are Lipschitz, that is, the local Lipschitz conditions (LLC) and global Lipschitz condition (GLC) are satisfied. In this work, we first demonstrate the importance of the above Lipschitzness assumptions on the state-of-the-art decentralized algorithms. First, by constructing a series of examples, we show that when the LLC on the local function gradient V fi's are not satisfied, a number of state-of-the-art decentralized algorithms diverge, even if the global Lipschitz condition (GLC) still holds. This observation brings out a fundamental theoretical issue of the existing decentralized algorithms---their convergence conditions are strictly stronger than centralized algorithms such as the gradient descent, which only requires the GLC. Our observation raises an important open question: How to design decentralized algorithms when the LLC, or even the GLC, is not satisfied? To address this question, we design two first-order algorithms, which are capable of computing stationary solutions of the original problem with neither the LLC nor the GLC condition. In particular, we show that the proposed algorithms converge sublinearly to a certain \epsilon -stationary solution, where the precise rate depends on various algorithmic and problem parameters. In particular, if the local function fi's are lower bounded Qth order polynomials, then the rate becomes 0(1/\epsilonQ-1) for Q \geq 2 (where the 0 notation hides some constants such as dependency on the network topology). Such a rate is tight for the special case of Q = 2 where each fi satisfies LLC. To our knowledge, this is the first attempt that studies decentralized nonconvex optimization problems with neither the LLC nor the GLC.
引用
收藏
页码:2879 / 2908
页数:30
相关论文
共 50 条
  • [41] Communication-efficient algorithms for decentralized and stochastic optimization
    Lan, Guanghui
    Lee, Soomin
    Zhou, Yi
    MATHEMATICAL PROGRAMMING, 2020, 180 (1-2) : 237 - 284
  • [42] Decentralized Dual Operator Splitting for Nonsmooth Composite Optimization
    Li, Huaqing
    Ding, Wentao
    Wang, Zheng
    Lu, Qingguo
    Li, Yongfu
    Xia, Dawen
    IEEE TRANSACTIONS ON NETWORK SCIENCE AND ENGINEERING, 2022, 9 (04): : 2084 - 2097
  • [43] Privacy-Preserving Push-Pull Method for Decentralized Optimization via State Decomposition
    Cheng, Huqiang
    Liao, Xiaofeng
    Li, Huaqing
    Lu, Qingguo
    Zhao, You
    IEEE TRANSACTIONS ON SIGNAL AND INFORMATION PROCESSING OVER NETWORKS, 2024, 10 : 513 - 526
  • [44] Decentralized Inexact Proximal Gradient Method With Network-Independent Stepsizes for Convex Composite Optimization
    Guo, Luyao
    Shi, Xinli
    Cao, Jinde
    Wang, Zihao
    IEEE TRANSACTIONS ON SIGNAL PROCESSING, 2023, 71 : 786 - 801
  • [45] A REDISTRIBUTED PROXIMAL BUNDLE METHOD FOR NONCONVEX OPTIMIZATION
    Hare, Warren
    Sagastizabal, Claudia
    SIAM JOURNAL ON OPTIMIZATION, 2010, 20 (05) : 2442 - 2473
  • [46] Distributed Nonconvex Optimization: Gradient-Free Iterations and ε-Globally Optimal Solution
    He, Zhiyu
    He, Jianping
    Chen, Cailian
    Guan, Xinping
    IEEE TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS, 2024, 11 (04): : 2239 - 2251
  • [47] On Distributed Nonconvex Optimization: Projected Subgradient Method for Weakly Convex Problems in Networks
    Chen, Shixiang
    Garcia, Alfredo
    Shahrampour, Shahin
    IEEE TRANSACTIONS ON AUTOMATIC CONTROL, 2022, 67 (02) : 662 - 675
  • [48] Asynchronous Speedup in Decentralized Optimization
    Even, Mathieu
    Hendrikx, Hadrien
    Massoulie, Laurent
    IEEE TRANSACTIONS ON AUTOMATIC CONTROL, 2025, 70 (03) : 1467 - 1482
  • [49] Decentralized Optimization with Edge Sampling
    Zhang, Chi
    Li, Qianxiao
    Zhao, Peilin
    PROCEEDINGS OF THE TWENTY-EIGHTH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE, 2019, : 658 - 664
  • [50] DECENTRALIZED EXACT COUPLED OPTIMIZATION
    Alghunaim, Sulaiman A.
    Yuan, Kun
    Sayed, Ali H.
    2017 55TH ANNUAL ALLERTON CONFERENCE ON COMMUNICATION, CONTROL, AND COMPUTING (ALLERTON), 2017, : 338 - 345