A Projection-free Algorithm for Constrained Stochastic Multi-level Composition Optimization

被引:0
作者
Xiao, Tesi [1 ]
Balasubramanian, Krishnakumar [1 ]
Ghadimi, Saeed [2 ]
机构
[1] Univ Calif Davis, Dept Stat, Davis, CA 95616 USA
[2] Univ Waterloo, Dept Management Sci, Waterloo, ON, Canada
来源
ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 35, NEURIPS 2022 | 2022年
基金
美国国家科学基金会; 加拿大自然科学与工程研究理事会;
关键词
APPROXIMATION METHOD; GRADIENT METHODS; MINIMIZATION;
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
We propose a projection-free conditional gradient-type algorithm for smooth stochastic multi-level composition optimization, where the objective function is a nested composition of T functions and the constraint set is a closed convex set. Our algorithm assumes access to noisy evaluations of the functions and their gradients, through a stochastic first-order oracle satisfying certain standard unbiasedness and second-moment assumptions. We show that the number of calls to the stochastic first-order oracle and the linear-minimization oracle required by the proposed algorithm, to obtain an epsilon-stationary solution, are of order O-T (epsilon(-2)) and O-T (epsilon(-3)) respectively, where OT hides constants in T. Notably, the dependence of these complexity bounds on epsilon and T are separate in the sense that changing one does not impact the dependence of the bounds on the other. For the case of T = 1, we also provide a high-probability convergence result that depends poly-logarithmically on the inverse confidence level. Moreover, our algorithm is parameter-free and does not require any (increasing) order of mini-batches to converge unlike the common practice in the analysis of stochastic conditional gradient-type algorithms.
引用
收藏
页数:13
相关论文
共 54 条
[1]  
Akhtar Z, 2022, Arxiv, DOI arXiv:2110.11721
[2]  
Astudillo Raul, 2021, Advances in Neural Information Processing Systems, V34
[3]   STOCHASTIC MULTILEVEL COMPOSITION OPTIMIZATION ALGORITHMS WITH LEVEL-INDEPENDENT CONVERGENCE RATES [J].
Balasubramanian, Krishnakumar ;
Ghadimi, Saeed ;
Nguyen, Anthony .
SIAM JOURNAL ON OPTIMIZATION, 2022, 32 (02) :519-544
[4]   Zeroth-Order Nonconvex Stochastic Optimization: Handling Constraints, High Dimensionality, and Saddle Points [J].
Balasubramanian, Krishnakumar ;
Ghadimi, Saeed .
FOUNDATIONS OF COMPUTATIONAL MATHEMATICS, 2022, 22 (01) :35-76
[5]   Linearly convergent away-step conditional gradient for non-strongly convex functions [J].
Beck, Amir ;
Shtern, Shimrit .
MATHEMATICAL PROGRAMMING, 2017, 164 (1-2) :1-27
[6]  
Blanchet J, 2017, Arxiv, DOI arXiv:1711.07564
[7]   Solving Stochastic Compositional Optimization is Nearly as Easy as Solving Stochastic Optimization [J].
Chen, Tianyi ;
Sun, Yuejiao ;
Yin, Wotao .
IEEE TRANSACTIONS ON SIGNAL PROCESSING, 2021, 69 :4937-4948
[8]   Minimal Variance Sampling with Provable Guarantees for Fast Training of Graph Neural Networks [J].
Cong, Weilin ;
Forsati, Rana ;
Kandemir, Mahmut ;
Mahdavi, Mehrdad .
KDD '20: PROCEEDINGS OF THE 26TH ACM SIGKDD INTERNATIONAL CONFERENCE ON KNOWLEDGE DISCOVERY & DATA MINING, 2020, :1393-1403
[9]   STOCHASTIC MODEL-BASED MINIMIZATION OF WEAKLY CONVEX FUNCTIONS [J].
Davis, Damek ;
Drusvyatskiy, Dmitriy .
SIAM JOURNAL ON OPTIMIZATION, 2019, 29 (01) :207-239
[10]   Statistical estimation of composite risk functionals and risk optimization problems [J].
Dentcheva, Darinka ;
Penev, Spiridon ;
Ruszczynski, Andrzej .
ANNALS OF THE INSTITUTE OF STATISTICAL MATHEMATICS, 2017, 69 (04) :737-760