Linearly convergent bilevel optimization with single-step inner methods

被引:1
作者
Suonpera, Ensio [1 ]
Valkonen, Tuomo [1 ,2 ]
机构
[1] Univ Helsinki, Dept Math & Stat, Helsinki, Finland
[2] Escuela Politec Nacl, ModeMat, Quito, Ecuador
基金
芬兰科学院;
关键词
Bilevel optimization; Nonsmooth; Inverse problems; Forward-backward;
D O I
10.1007/s10589-023-00527-7
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 propose a new approach to solving bilevel optimization problems, intermediate between solving full-system optimality conditions with a Newton-type approach, and treating the inner problem as an implicit function. The overall idea is to solve the full-system optimality conditions, but to precondition them to alternate between taking steps of simple conventional methods for the inner problem, the adjoint equation, and the outer problem. While the inner objective has to be smooth, the outer objective may be nonsmooth subject to a prox-contractivity condition. We prove linear convergence of the approach for combinations of gradient descent and forward-backward splitting with exact and inexact solution of the adjoint equation. We demonstrate good performance on learning the regularization parameter for anisotropic total variation image denoising, and the convolution kernel for image deconvolution.
引用
收藏
页码:571 / 610
页数:40
相关论文
共 55 条
  • [1] Aussel D., 2017, GEN NASH EQUILIBRIUM, DOI DOI 10.1007/978-981-10-4774-9
  • [2] Directional Necessary Optimality Conditions for Bilevel Programs
    Bai, Kuang
    Ye, Jane J.
    [J]. MATHEMATICS OF OPERATIONS RESEARCH, 2022, 47 (02) : 1169 - 1191
  • [3] AN EXPLICIT SOLUTION TO THE MULTILEVEL PROGRAMMING PROBLEM
    BARD, JF
    FALK, JE
    [J]. COMPUTERS & OPERATIONS RESEARCH, 1982, 9 (01) : 77 - 100
  • [4] Solving bilevel programs with the KKT-approach
    Bouza Allende, Gemayqzel
    Still, Georg
    [J]. MATHEMATICAL PROGRAMMING, 2013, 138 (1-2) : 309 - 332
  • [5] Calatroni C., 2017, Variational Methods: In Imaging and Geometric Control, V18, P252, DOI [DOI 10.1515/9783110430394, 10.1515/9783110430394-008, DOI 10.1515/9783110430394-008]
  • [6] Calatroni L., 2014, P 26 IFIP TC 7 C SYS
  • [7] Optimality Conditions for Bilevel Imaging Learning Problems with Total Variation Regularization
    Carlos De los Reyes, Juan
    Villacis, David
    [J]. SIAM JOURNAL ON IMAGING SCIENCES, 2022, 15 (04): : 1646 - 1689
  • [8] IMAGE DENOISING: LEARNING THE NOISE MODEL VIA NONSMOOTH PDE-CONSTRAINED OPTIMIZATION
    Carlos De los Reyes, Juan
    Schoenlieb, Carola-Bibiane
    [J]. INVERSE PROBLEMS AND IMAGING, 2013, 7 (04) : 1183 - 1214
  • [9] Learning Consistent Discretizations of the Total Variation
    Chambolle, Antonin
    Pock, Thomas
    [J]. SIAM JOURNAL ON IMAGING SCIENCES, 2021, 14 (02): : 778 - 813
  • [10] A First-Order Primal-Dual Algorithm for Convex Problems with Applications to Imaging
    Chambolle, Antonin
    Pock, Thomas
    [J]. JOURNAL OF MATHEMATICAL IMAGING AND VISION, 2011, 40 (01) : 120 - 145