FedNS: A Fast Sketching Newton-Type Algorithm for Federated Learning

被引:0
|
作者
Li, Jian [1 ]
Liu, Yong [2 ]
Wang, Weiping [1 ]
机构
[1] Chinese Acad Sci, Inst Informat Engn, Beijing, Peoples R China
[2] Renmin Univ China, Gaoling Sch Artificial Intelligence, Beijing, Peoples R China
基金
中国博士后科学基金; 中国国家自然科学基金;
关键词
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Recent Newton-type federated learning algorithms have demonstrated linear convergence with respect to the communication rounds. However, communicating Hessian matrices is often unfeasible due to their quadratic communication complexity. In this paper, we introduce a novel approach to tackle this issue while still achieving fast convergence rates. Our proposed method, named as Federated Newton Sketch methods (FedNS), approximates the centralized Newton's method by communicating the sketched square-root Hessian instead of the exact Hessian. To enhance communication efficiency, we reduce the sketch size to match the effective dimension of the Hessian matrix. We provide convergence analysis based on statistical learning for the federated Newton sketch approaches. Specifically, our approaches reach super-linear convergence rates w.r.t. the communication rounds for the first time. We validate the effectiveness of our algorithms through various experiments, which coincide with our theoretical findings.
引用
收藏
页码:13509 / 13517
页数:9
相关论文
共 50 条
  • [41] On the construction of probabilistic Newton-type algorithms
    Wills, Adrian G.
    Schon, Thomas B.
    2017 IEEE 56TH ANNUAL CONFERENCE ON DECISION AND CONTROL (CDC), 2017,
  • [42] Newton-type multilevel optimization method
    Ho, Chin Pang
    Kocvara, Michal
    Parpas, Panos
    OPTIMIZATION METHODS & SOFTWARE, 2022, 37 (01): : 45 - 78
  • [43] On Newton-type methods with cubic convergence
    Homeier, HHH
    JOURNAL OF COMPUTATIONAL AND APPLIED MATHEMATICS, 2005, 176 (02) : 425 - 432
  • [44] Newton-Type Methods: A Broader View
    Izmailov, A. F.
    Solodov, M. V.
    JOURNAL OF OPTIMIZATION THEORY AND APPLICATIONS, 2015, 164 (02) : 577 - 620
  • [45] Newton-type methods for stochastic programming
    Chen, X
    MATHEMATICAL AND COMPUTER MODELLING, 2000, 31 (10-12) : 89 - 98
  • [46] On fractional biparameterized Newton-type inequalities
    Saleh, Wedad
    Lakhdari, Abdelghani
    Abdeljawad, Thabet
    Meftah, Badreddine
    JOURNAL OF INEQUALITIES AND APPLICATIONS, 2023, 2023 (01):
  • [47] Newton-type methods in array processing
    Selva, J
    IEEE SIGNAL PROCESSING LETTERS, 2004, 11 (02) : 104 - 107
  • [48] Newton-type methods for quasidifferentiable equations
    Zhang, LW
    Xia, ZQ
    JOURNAL OF OPTIMIZATION THEORY AND APPLICATIONS, 2001, 108 (02) : 439 - 456
  • [49] Newton-Type Methods for Quasidifferentiable Equations
    L. W. Zhang
    Z. Q. Xia
    Journal of Optimization Theory and Applications, 2001, 108 : 439 - 456
  • [50] On the convergence of an inexact Newton-type method
    Zhou, Guanglu
    Qi, Liqun
    OPERATIONS RESEARCH LETTERS, 2006, 34 (06) : 647 - 652