A stochastic subspace approach to gradient-free optimization in high dimensions

被引:15
|
作者
Kozak, David [1 ,2 ]
Becker, Stephen [3 ]
Doostan, Alireza [4 ]
Tenorio, Luis [1 ]
机构
[1] Solea Energy, Overland Pk, KS 66210 USA
[2] Colorado Sch Mines, Dept Appl Math & Stat, Golden, CO 80401 USA
[3] Univ Colorado, Dept Appl Math, Boulder, CO 80309 USA
[4] Univ Colorado, Dept Aerosp Engn Sci, Boulder, CO 80309 USA
基金
美国国家科学基金会;
关键词
Randomized methods; Gradient-free; Gaussian processes; Stochastic gradients; INVERSE PROBLEMS; CONVEX-PROGRAMS; DESCENT; ALGORITHM; MINIMIZATION; UNCERTAINTY; PERFORMANCE; COMPLEXITY; FLOW;
D O I
10.1007/s10589-021-00271-w
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
We present a stochastic descent algorithm for unconstrained optimization that is particularly efficient when the objective function is slow to evaluate and gradients are not easily obtained, as in some PDE-constrained optimization and machine learning problems. The algorithm maps the gradient onto a low-dimensional random subspace of dimension l at each iteration, similar to coordinate descent but without restricting directional derivatives to be along the axes. Without requiring a full gradient, this mapping can be performed by computing l directional derivatives (e.g., via forward-mode automatic differentiation). We give proofs for convergence in expectation under various convexity assumptions as well as probabilistic convergence results under strong-convexity. Our method provides a novel extension to the well-known Gaussian smoothing technique to descent in subspaces of dimension greater than one, opening the doors to new analysis of Gaussian smoothing when more than one directional derivative is used at each iteration. We also provide a finite-dimensional variant of a special case of the Johnson-Lindenstrauss lemma. Experimentally, we show that our method compares favorably to coordinate descent, Gaussian smoothing, gradient descent and BFGS (when gradients are calculated via forward-mode automatic differentiation) on problems from the machine learning and shape optimization literature.
引用
收藏
页码:339 / 368
页数:30
相关论文
共 50 条
  • [1] A stochastic subspace approach to gradient-free optimization in high dimensions
    David Kozak
    Stephen Becker
    Alireza Doostan
    Luis Tenorio
    Computational Optimization and Applications, 2021, 79 : 339 - 368
  • [2] An adaptive Bayesian approach to gradient-free global optimization
    Yu, Jianneng
    Morozov, Alexandre, V
    NEW JOURNAL OF PHYSICS, 2024, 26 (02):
  • [3] Accelerating Groundwater Data Assimilation With a Gradient-Free Active Subspace Method
    Yan, Hengnian
    Hao, Chenyu
    Zhang, Jiangjiang
    Illman, Walter A.
    Lin, Guang
    Zeng, Lingzao
    WATER RESOURCES RESEARCH, 2021, 57 (12)
  • [4] Ensemble Neural Networks (ENN): A gradient-free stochastic method
    Chen, Yuntian
    Chang, Haibin
    Meng, Jin
    Zhang, Dongxiao
    NEURAL NETWORKS, 2019, 110 : 170 - 185
  • [5] Mixed Simultaneous Perturbation Stochastic Approximation for Gradient-Free Optimization with Noisy Measurements
    Wang, Long
    Zhu, Jingyi
    Spall, James C.
    2018 ANNUAL AMERICAN CONTROL CONFERENCE (ACC), 2018, : 3774 - 3779
  • [6] Gradient-free distributed online optimization in networks
    Liu, Yuhang
    Zhao, Wenxiao
    Zhang, Nan
    Lv, Dongdong
    Zhang, Shuai
    CONTROL THEORY AND TECHNOLOGY, 2025,
  • [7] Gradient-Free Aeroacoustic Shape Optimization Using Large Eddy Simulation
    Hamedi, Mohsen
    Vermeire, Brian
    AIAA JOURNAL, 2025,
  • [8] High-Dimensional Nonlinear Multi-Fidelity Model with Gradient-Free Active Subspace Method
    Liu, Bangde
    Lin, Guang
    COMMUNICATIONS IN COMPUTATIONAL PHYSICS, 2020, 28 (05) : 1937 - 1969
  • [9] High-fidelity gradient-free optimization of low-pressure turbine cascades
    Aubry, Anthony
    Karbasian, Hamid R.
    Vermeire, Brian C.
    COMPUTERS & FLUIDS, 2022, 248
  • [10] Gradient-free aerodynamic shape optimization using Large Eddy Simulation
    Karbasian, Hamid R.
    Vermeire, Brian C.
    COMPUTERS & FLUIDS, 2022, 232