Stochastic Optimization with Heavy-Tailed Noise via Accelerated Gradient Clipping

被引:0
|
作者
Gorbunov, Eduard [1 ,2 ]
Danilova, Marina [3 ]
Gasnikov, Alexander [1 ,2 ]
机构
[1] MIPT, Dolgoprudnyi, Russia
[2] HSE, Moscow, Russia
[3] RAS Russia, ICS, Moscow, Russia
关键词
APPROXIMATION ALGORITHMS; COMPOSITE OPTIMIZATION; INEQUALITIES; 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
In this paper, we propose a new accelerated stochastic first-order method called clipped-SSTM for smooth convex stochastic optimization with heavy-tailed distributed noise in stochastic gradients and derive the first high-probability complexity bounds for this method closing the gap in the theory of stochastic optimization with heavy-tailed noise. Our method is based on a special variant of accelerated Stochastic Gradient Descent (SGD) and clipping of stochastic gradients. We extend our method to the strongly convex case and prove new complexity bounds that outperform state-of-the-art results in this case. Finally, we extend our proof technique and derive the first non-trivial high-probability complexity bounds for SGD with clipping without light-tails assumption on the noise.
引用
收藏
页数:12
相关论文
共 50 条
  • [1] Algorithms with Gradient Clipping for Stochastic Optimization with Heavy-Tailed Noise
    Danilova, M.
    [J]. DOKLADY MATHEMATICS, 2023, 108 (SUPPL 2) : S248 - S256
  • [2] Algorithms with Gradient Clipping for Stochastic Optimization with Heavy-Tailed Noise
    M. Danilova
    [J]. Doklady Mathematics, 2023, 108 : S248 - S256
  • [3] Stochastic PDEs with heavy-tailed noise
    Chong, Carsten
    [J]. STOCHASTIC PROCESSES AND THEIR APPLICATIONS, 2017, 127 (07) : 2262 - 2280
  • [4] Breaking the Heavy-Tailed Noise Barrier in Stochastic Optimization Problems
    Puchkin, Nikita
    Gorbunov, Eduard
    Kutuzov, Nikolay
    Gasnikov, Alexander
    [J]. INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE AND STATISTICS, VOL 238, 2024, 238
  • [5] First Exit Time Analysis of Stochastic Gradient Descent Under Heavy-Tailed Gradient Noise
    Thanh Huy Nguyen
    Simsekli, Umut
    Gurbuzbalaban, Mert
    Richard, Gael
    [J]. ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 32 (NIPS 2019), 2019, 32
  • [6] Gradient-free methods for non-smooth convex stochastic optimization with heavy-tailed noise on convex compact
    Nikita Kornilov
    Alexander Gasnikov
    Pavel Dvurechensky
    Darina Dvinskikh
    [J]. Computational Management Science, 2023, 20
  • [7] Gradient-free methods for non-smooth convex stochastic optimization with heavy-tailed noise on convex compact
    Kornilov, Nikita
    Gasnikov, Alexander
    Dvurechensky, Pavel
    Dvinskikh, Darina
    [J]. COMPUTATIONAL MANAGEMENT SCIENCE, 2023, 20 (01)
  • [8] Efficient private SCO for heavy-tailed data via averaged clipping
    Jin, Chenhan
    Zhou, Kaiwen
    Han, Bo
    Cheng, James
    Zeng, Tieyong
    [J]. MACHINE LEARNING, 2024,
  • [9] Algorithmic Stability of Heavy-Tailed Stochastic Gradient Descent on Least Squares
    Raj, Anant
    Barsbey, Melih
    Gurbuzbalaban, Mert
    Zhu, Lingjiong
    Simsekli, Umut
    [J]. INTERNATIONAL CONFERENCE ON ALGORITHMIC LEARNING THEORY, VOL 201, 2023, 201 : 1292 - 1342
  • [10] Optimization of M-of-N detectors in heavy-tailed noise
    Abraham, D. A.
    [J]. OCEANS 2018 MTS/IEEE CHARLESTON, 2018,