Performance of noisy three-step accelerated first-order optimization algorithms for strongly convex quadratic problems

被引:0
|
作者
Samuelson, Samantha [1 ]
Mohammadi, Hesameddin [1 ]
Jovanovic, Mihailo R. [1 ]
机构
[1] Univ Southern Calif, Dept Elect & Comp Engn, Los Angeles, CA 90089 USA
来源
2023 62ND IEEE CONFERENCE ON DECISION AND CONTROL, CDC | 2023年
关键词
Convex optimization; gradient descent; heavy-ball method; Nesterov's accelerated algorithms; noisy gradients; performance tradeoffs;
D O I
10.1109/CDC49753.2023.10383581
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
We study the class of first-order algorithms in which the optimization variable is updated using information from three previous iterations. While two-step momentum algorithms akin to heavy-ball and Nesterov's accelerated methods achieve the optimal convergence rate, it is an open question if the three-step momentum method can offer advantages for problems in which exact gradients are not available. For strongly convex quadratic problems, we identify algorithmic parameters which achieve the optimal convergence rate and examine how additional momentum terms affects the tradeoffs between acceleration and noise amplification. Our results suggest that for parameters that optimize the convergence rate, introducing additional momentum terms does not provide improvement in variance amplification relative to standard accelerated algorithms.
引用
收藏
页码:1300 / 1305
页数:6
相关论文
共 13 条
  • [1] Variance amplification of accelerated first-order algorithms for strongly convex quadratic optimization problems
    Mohammadi, Hesameddin
    Razaviyayn, Meisam
    Jovanovic, Mihailo R.
    2018 IEEE CONFERENCE ON DECISION AND CONTROL (CDC), 2018, : 5753 - 5758
  • [2] Robustness of Accelerated First-Order Algorithms for Strongly Convex Optimization Problems
    Mohammadi, Hesameddin
    Razaviyayn, Meisam
    Jovanovic, Mihailo R.
    IEEE TRANSACTIONS ON AUTOMATIC CONTROL, 2021, 66 (06) : 2480 - 2495
  • [3] Performance of noisy higher-order accelerated gradient flow dynamics for strongly convex quadratic optimization problems
    Samuelson, Samantha
    Mohammadi, Hesameddin
    Jovanovic, Mihailo R.
    2023 AMERICAN CONTROL CONFERENCE, ACC, 2023, : 3839 - 3844
  • [4] Performance of noisy Nesterov's accelerated method for strongly convex optimization problems
    Mohammadi, Hesameddin
    Razaviyayn, Meisam
    Jovanovic, Mihailo R.
    2019 AMERICAN CONTROL CONFERENCE (ACC), 2019, : 3426 - 3431
  • [5] Accelerated First-Order Optimization Algorithms for Machine Learning
    Li, Huan
    Fang, Cong
    Lin, Zhouchen
    PROCEEDINGS OF THE IEEE, 2020, 108 (11) : 2067 - 2082
  • [6] First-Order Algorithms for Convex Optimization with Nonseparable Objective and Coupled Constraints
    Gao X.
    Zhang S.-Z.
    Zhang, Shu-Zhong (zhangs@umn.edu), 1600, Springer Science and Business Media Deutschland GmbH (05): : 131 - 159
  • [7] From differential equation solvers to accelerated first-order methods for convex optimization
    Hao Luo
    Long Chen
    Mathematical Programming, 2022, 195 : 735 - 781
  • [8] ACCELERATED FIRST-ORDER METHODS FOR CONVEX OPTIMIZATION WITH LOCALLY LIPSCHITZ CONTINUOUS GRADIENT
    Lu, Zhaosong
    Mei, Sanyou
    SIAM JOURNAL ON OPTIMIZATION, 2023, 33 (03) : 2275 - 2310
  • [9] From differential equation solvers to accelerated first-order methods for convex optimization
    Luo, Hao
    Chen, Long
    MATHEMATICAL PROGRAMMING, 2022, 195 (1-2) : 735 - 781
  • [10] FOM - a MATLAB toolbox of first-order methods for solving convex optimization problems
    Beck, Amir
    Guttmann-Beck, Nili
    OPTIMIZATION METHODS & SOFTWARE, 2019, 34 (01) : 172 - 193