Consistency and fluctuations for stochastic gradient Langevin dynamics

被引:0
|
作者
机构
[1] Teh, Yee Whye
[2] Thiery, Alexandre H.
[3] Vollmer, Sebastian J.
来源
| 1600年 / Microtome Publishing卷 / 17期
关键词
Mean square error - Monte Carlo methods - Chains - Iterative methods - Markov processes - Stochastic systems - Dynamics;
D O I
暂无
中图分类号
学科分类号
摘要
Applying standard Markov chain Monte Carlo (MCMC) algorithms to large data sets is computationally expensive. Both the calculation of the acceptance probability and the creation of informed proposals usually require an iteration through the whole data set. The recently proposed stochastic gradient Langevin dynamics (SGLD) method circumvents this problem by generating proposals which are only based on a subset of the data, by skipping the accept-reject step and by using decreasing step-sizes sequence (δm)m≥0. We provide in this article a rigorous mathematical framework for analysing this algorithm. We prove that, under verifiable assumptions, the algorithm is consistent, satisfies a central limit theorem (CLT) and its asymptotic bias-variance decomposition can be characterized by an explicit functional of the step-sizes sequence (δm)m≥0. We leverage this analysis to give practical recommendations for the notoriously dificult tuning of this algorithm: it is asymptotically optimal to use a step-size sequence of the type δm = mm-1/3, leading to an algorithm whose mean squared error (MSE) decreases at rate O(m-1/3). © 2016 Yee Whye Teh, Alexandre H. Thiery, and Sebastian J. Vollmer.
引用
收藏
相关论文
共 50 条
  • [1] Consistency and Fluctuations For Stochastic Gradient Langevin Dynamics
    Teh, Yee Whye
    Thiery, Alexandre H.
    Vollmer, Sebastian J.
    JOURNAL OF MACHINE LEARNING RESEARCH, 2016, 17
  • [2] Stochastic Gradient Langevin Dynamics with Variance Reduction
    Huang, Zhishen
    Becker, Stephen
    2021 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS (IJCNN), 2021,
  • [3] The promises and pitfalls of Stochastic Gradient Langevin Dynamics
    Brosse, Nicolas
    Moulines, Eric
    Durmus, Alain
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 31 (NIPS 2018), 2018, 31
  • [4] Variance Reduction in Stochastic Gradient Langevin Dynamics
    Dubey, Avinava
    Reddi, Sashank J.
    Poczos, Barnabas
    Smola, Alexander J.
    Xing, Eric P.
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 29 (NIPS 2016), 2016, 29
  • [5] Stochastic gradient Langevin dynamics with adaptive drifts
    Kim, Sehwan
    Song, Qifan
    Liang, Faming
    JOURNAL OF STATISTICAL COMPUTATION AND SIMULATION, 2022, 92 (02) : 318 - 336
  • [6] Evaluating and Diagnosing Convergence for Stochastic Gradient Langevin Dynamics
    Hernandez, Sergio
    Luis Lopez, Juan
    2021 40TH INTERNATIONAL CONFERENCE OF THE CHILEAN COMPUTER SCIENCE SOCIETY (SCCC), 2021,
  • [7] Characterizing Membership Privacy in Stochastic Gradient Langevin Dynamics
    Wu, Bingzhe
    Chen, Chaochao
    Zhao, Shiwan
    Chen, Cen
    Yao, Yuan
    Sun, Guangyu
    Li Wang
    Zhang, Xiaolu
    Zhou, Jun
    THIRTY-FOURTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE, THE THIRTY-SECOND INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE CONFERENCE AND THE TENTH AAAI SYMPOSIUM ON EDUCATIONAL ADVANCES IN ARTIFICIAL INTELLIGENCE, 2020, 34 : 6372 - 6379
  • [8] Low-Precision Stochastic Gradient Langevin Dynamics
    Zhang, Ruqi
    Wilson, Andrew Gordon
    De Sa, Christopher
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 162, 2022,
  • [9] Stochastic Gradient Langevin Dynamics for Massive MIMO Detection
    Wu, Zhiwen
    Li, Hui
    IEEE COMMUNICATIONS LETTERS, 2022, 26 (05) : 1062 - 1065
  • [10] Decentralized Stochastic Gradient Langevin Dynamics and Hamiltonian Monte Carlo
    Gurbuzbalaban, Mert
    Gao, Xuefeng
    Hu, Yuanhan
    Zhu, Lingjiong
    JOURNAL OF MACHINE LEARNING RESEARCH, 2021, 22