A Variable Sample-Size Stochastic Quasi-Newton Method for Smooth and Nonsmooth Stochastic Convex Optimization

被引:6
|
作者
Jalilzadeh, Afrooz [1 ]
Nedic, Angelia [2 ]
Shanbhag, Uday, V [1 ]
Yousefian, Farzad [3 ]
机构
[1] Penn State Univ, Ind & Mfg Engn, State Coll, PA 16802 USA
[2] Arizona State Univ, Sch Elect Comp & Energy Engn, Tempe, AZ 85287 USA
[3] Oklahoma State Univ, Sch Ind Engn & Management, Stillwater, OK 74078 USA
基金
美国国家科学基金会;
关键词
stochastic optimization; convex optimization; quasi-Newton methods; LINE SEARCH; CONVERGENCE; ALGORITHMS;
D O I
10.1287/moor.2021.1147
中图分类号
C93 [管理学]; O22 [运筹学];
学科分类号
070105 ; 12 ; 1201 ; 1202 ; 120202 ;
摘要
Classical theory for quasi-Newton schemes has focused on smooth, deterministic, unconstrained optimization, whereas recent forays into stochastic convex optimization have largely resided in smooth, unconstrained, and strongly convex regimes. Naturally, there is a compelling need to address nonsmoothness, the lack of strong convexity, and the presence of constraints. Accordingly, this paper presents a quasi-Newton framework that can process merely convex and possibly nonsmooth (but smoothable) stochastic convex problems. We propose a framework that combines iterative smoothing and regularization with a variance-reduced scheme reliant on using an increasing sample size of gradients. We make the following contributions. (i) We develop a regularized and smoothed variable sample-size BFGS update (rsL-BFGS) that generates a sequence of Hessian approximations and can accommodate nonsmooth convex objectives by utilizing iterative regularization and smoothing. (ii) In strongly convex regimes with state-dependent noise, the proposed variable sample-size stochastic quasi-Newton (VS-SQN) scheme admits a nonasymptotic linear rate of convergence, whereas the oracle complexity of computing an epsilon-solution is O(k(m+1)/epsilon), where K denotes the condition number and m >= 1. In nonsmooth (but smooth-able) regimes, using Moreau smoothing retains the linear convergence rate for the resulting smoothed VS-SQN (or sVS-SQN) scheme. Notably, the nonsmooth regime allows for accommodating convex constraints. To contend with the possible unavailability of Lipschitzian and strong convexity parameters, we also provide sublinear rates for diminishing step-length variants that do not rely on the knowledge of such parameters. (iii) In merely convex but smooth settings, the regularized VS-SQN scheme rVS-SQN displays a rate of O(1/k((1-epsilon))) with an oracle complexity of O(1/epsilon(3)). When the smoothness requirements are weakened, the rate for the regularized and smoothed VS-SQN scheme rsVS-SQN worsens to O(k(-1/3)). Such statements allow for a state-dependent noise assumption under a quadratic growth property on the objective. To the best of our knowledge, the rate results are among the first available rates for QN methods in nonsmooth regimes. Preliminary numerical evidence suggests that the schemes compare well with accelerated gradient counterparts on selected problems in stochastic optimization and machine learning with significant benefits in ill-conditioned regimes.
引用
收藏
页码:690 / 719
页数:30
相关论文
共 50 条
  • [1] A Variable Sample-size Stochastic Quasi-Newton Method for Smooth and Nonsmooth Stochastic Convex Optimization
    Jalilzadeh, Afrooz
    Nedic, Angelia
    Shanbhag, Uday V.
    Yousefian, Farzad
    2018 IEEE CONFERENCE ON DECISION AND CONTROL (CDC), 2018, : 4097 - 4102
  • [2] STOCHASTIC QUASI-NEWTON METHOD FOR NONCONVEX STOCHASTIC OPTIMIZATION
    Wang, Xiao
    Ma, Shiqian
    Goldfarb, Donald
    Liu, Wei
    SIAM JOURNAL ON OPTIMIZATION, 2017, 27 (02) : 927 - 956
  • [3] A stochastic extra-step quasi-Newton method for nonsmooth nonconvex optimization
    Minghan Yang
    Andre Milzarek
    Zaiwen Wen
    Tong Zhang
    Mathematical Programming, 2022, 194 : 257 - 303
  • [4] A stochastic extra-step quasi-Newton method for nonsmooth nonconvex optimization
    Yang, Minghan
    Milzarek, Andre
    Wen, Zaiwen
    Zhang, Tong
    MATHEMATICAL PROGRAMMING, 2022, 194 (1-2) : 257 - 303
  • [5] A single timescale stochastic quasi-Newton method for stochastic optimization
    Wang, Peng
    Zhu, Detong
    INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS, 2023, 100 (12) : 2196 - 2216
  • [6] A stochastic quasi-Newton method for simulation response optimization
    Kao, C
    Chen, SP
    EUROPEAN JOURNAL OF OPERATIONAL RESEARCH, 2006, 173 (01) : 30 - 46
  • [7] Quasi-Newton methods for stochastic optimization
    Levy, MN
    Trosset, MW
    Kincaid, RR
    ISUMA 2003: FOURTH INTERNATIONAL SYMPOSIUM ON UNCERTAINTY MODELING AND ANALYSIS, 2003, : 304 - 309
  • [8] Asynchronous Stochastic Quasi-Newton MCMC for Non-Convex Optimization
    Simsekli, Umut
    Yildiz, Cagatay
    Thanh Huy Nguyen
    Richard, Gael
    Cemgil, A. Taylan
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 80, 2018, 80
  • [9] A STOCHASTIC QUASI-NEWTON METHOD FOR LARGE-SCALE OPTIMIZATION
    Byrd, R. H.
    Hansen, S. L.
    Nocedal, Jorge
    Singer, Y.
    SIAM JOURNAL ON OPTIMIZATION, 2016, 26 (02) : 1008 - 1031
  • [10] Stochastic proximal quasi-Newton methods for non-convex composite optimization
    Wang, Xiaoyu
    Wang, Xiao
    Yuan, Ya-xiang
    OPTIMIZATION METHODS & SOFTWARE, 2019, 34 (05): : 922 - 948