Non-asymptotic Analysis of Biased Stochastic Approximation Scheme

被引:0
|
作者
Karimi, Belhal [1 ]
Miasojedow, Blazej [2 ]
Moulines, Eric [1 ]
Wai, Hoi-To [3 ]
机构
[1] Ecole Polytechn, CMAP, Palaiseau, France
[2] Univ Warsaw, Fac Math Informat & Mech, Warsaw, Poland
[3] Chinese Univ Hong Kong, Dept SEEM, Hong Kong, Peoples R China
来源
CONFERENCE ON LEARNING THEORY, VOL 99 | 2019年 / 99卷
关键词
biased stochastic approximation; state-dependent Markov chain; non-convex optimization; policy gradient; online expectation-maximization; GRADIENT; OPTIMIZATION; CONVERGENCE; ALGORITHMS;
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
Stochastic approximation (SA) is a key method used in statistical learning. Recently, its non-asymptotic convergence analysis has been considered in many papers. However, most of the prior analyses are made under restrictive assumptions such as unbiased gradient estimates and convex objective function, which significantly limit their applications to sophisticated tasks such as online and reinforcement learning. These restrictions are all essentially relaxed in this work. In particular, we analyze a general SA scheme to minimize a non-convex, smooth objective function. We consider update procedure whose drift term depends on a state-dependent Markov chain and the mean field is not necessarily of gradient type, covering approximate second-order method and allowing asymptotic bias for the one-step updates. We illustrate these settings with the online EM algorithm and the policy-gradient method for average reward maximization in reinforcement learning.
引用
收藏
页数:31
相关论文
共 50 条
  • [1] Non-Asymptotic Analysis of Stochastic Approximation Algorithms for Streaming Data
    Godichon-Baggioni, Antoine
    Werge, Nicklas
    Wintenberger, Olivier
    ESAIM-PROBABILITY AND STATISTICS, 2023, 27 : 482 - 514
  • [2] Non-asymptotic error bounds for constant stepsize stochastic approximation for tracking mobile agents
    Kumar, Bhumesh
    Borkar, Vivek
    Shetty, Akhil
    MATHEMATICS OF CONTROL SIGNALS AND SYSTEMS, 2019, 31 (04) : 589 - 614
  • [3] Normal Approximation for Stochastic Gradient Descent via Non-Asymptotic Rates of Martingale CLT
    Anastasiou, Andreas
    Balasubramanian, Krishnakumar
    Erdogdu, Murat A.
    CONFERENCE ON LEARNING THEORY, VOL 99, 2019, 99
  • [4] Optimal non-asymptotic analysis of the Ruppert-Polyak averaging stochastic algorithm
    Gadat, Sebastien
    Panloup, Fabien
    STOCHASTIC PROCESSES AND THEIR APPLICATIONS, 2023, 156 : 312 - 348
  • [5] Non-asymptotic confidence bounds for the optimal value of a stochastic program
    Guigues, Vincent
    Juditsky, Anatoli
    Nemirovski, Arkadi
    OPTIMIZATION METHODS & SOFTWARE, 2017, 32 (05) : 1033 - 1058
  • [6] Non-asymptotic Analysis of Stochastic Methods for Non-Smooth Non-Convex Regularized Problems
    Xu, Yi
    Jin, Rong
    Yang, Tianbao
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 32 (NIPS 2019), 2019, 32
  • [7] Sharp non-asymptotic concentration inequalities for the approximation of the invariant distribution of a diffusion
    Honore, Igor
    STOCHASTIC PROCESSES AND THEIR APPLICATIONS, 2020, 130 (04) : 2127 - 2158
  • [8] Non-asymptotic Gaussian estimates for the recursive approximation of the invariant distribution of a diffusion
    Honor, I
    Menozzi, S.
    Pages, G.
    ANNALES DE L INSTITUT HENRI POINCARE-PROBABILITES ET STATISTIQUES, 2020, 56 (03): : 1559 - 1605
  • [9] A Non-asymptotic Analysis of Non-parametric Temporal-Difference Learning
    Berthier, Eloise
    Kobeissi, Ziad
    Bach, Francis
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 35 (NEURIPS 2022), 2022,
  • [10] Almost Sure Convergence and Non-Asymptotic Concentration Bounds for Stochastic Mirror Descent Algorithm
    Paul, Anik Kumar
    Mahindrakar, Arun D.
    Kalaimani, Rachel K.
    IEEE CONTROL SYSTEMS LETTERS, 2024, 8 : 2397 - 2402