Convergence Analysis of ODE Models for Accelerated First-Order Methods via Positive Semidefinite Kernels

被引:0
作者
Kim, Jungbin [1 ]
Yang, Insoon [1 ]
机构
[1] Seoul Natl Univ, Seoul, South Korea
来源
ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 36 (NEURIPS 2023) | 2023年
基金
新加坡国家研究基金会;
关键词
WORST-CASE PERFORMANCE; OPTIMIZATION ALGORITHMS; GRADIENT-METHOD; CONVEX;
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 novel methodology that systematically analyzes ordinary differential equation (ODE) models for first-order optimization methods by converting the task of proving convergence rates into verifying the positive semidefiniteness of specific Hilbert-Schmidt integral operators. Our approach is based on the performance estimation problems (PEP) introduced by Drori and Teboulle [8]. Unlike previous works on PEP, which rely on finite-dimensional linear algebra, we use tools from functional analysis. Using the proposed method, we establish convergence rates of various accelerated gradient flow models, some of which are new. As an immediate consequence of our framework, we show a correspondence between minimizing function values and minimizing gradient norms.
引用
收藏
页数:13
相关论文
共 49 条
[31]  
Scieur D., 2017, Advances in Neural Information Processing Systems, V30
[32]   THE CONNECTIONS BETWEEN LYAPUNOV FUNCTIONS FOR SOME OPTIMIZATION ALGORITHMS AND DIFFERENTIAL EQUATIONS [J].
Serna, J. M. Sanz ;
Zygalakis, K. C. .
SIAM JOURNAL ON NUMERICAL ANALYSIS, 2021, 59 (03) :1542-1565
[33]  
Shi B, 2019, ADV NEUR IN, V32
[34]   Understanding the acceleration phenomenon via high-resolution differential equations [J].
Shi, Bin ;
Du, Simon S. ;
Jordan, Michael, I ;
Su, Weijie J. .
MATHEMATICAL PROGRAMMING, 2022, 195 (1-2) :79-148
[35]  
Stein E.M., 2009, Real analysis: measure theory, integration, and Hilbert spaces
[36]  
Su WJ, 2016, J MACH LEARN RES, V17
[37]  
Suh J. J., 2022, P MACHINE LEARNING R, P20640
[38]  
Sun BY, 2020, IEEE DECIS CONTR P, P4237, DOI 10.1109/CDC42340.2020.9304444
[39]  
Taylor A., 2019, En Conference on learning theory, P2934
[40]  
Taylor A, 2018, PR MACH LEARN RES, V80