Uniform-in-Time Weak Error Analysis for Stochastic Gradient Descent Algorithms via Diffusion Approximation
Uniform-in-Time Weak Error Analysis for Stochastic Gradient Descent Algorithms via Diffusion Approximation
复制标题
通过扩散近似对随机梯度下降算法进行均匀时间弱误差分析
DOI:
10.4310/cms.2020.v18.n1.a7
复制
发表时间:
2019-02
影响因子:
1
通讯作者:
Yulong Lu
中科院分区:
文献类型:
--
作者:
Yuanyuan Feng;Tingran Gao;Lei Li;Jian-Guo Liu;Yulong Lu
Diffusion approximation provides weak approximation for stochastic gradient descent algorithms in a finite time horizon. In this paper, we introduce new tools motivated by the backward error analysis of numerical stochastic differential equations into the theoretical framework of diffusion approximation, extending the validity of the weak approximation from finite to infinite time horizon. The new techniques developed in this paper enable us to characterize the asymptotic behavior of constant-step-size SGD algorithms for strongly convex objective functions, a goal previously unreachable within the diffusion approximation framework. Our analysis builds upon a truncated formal power expansion of the solution of a stochastic modified equation arising from diffusion approximation, where the main technical ingredient is a uniform-in-time weak error bound controlling the long-term behavior of the expansion coefficient functions near the global minimum. We expect these new techniques to greatly expand the range of applicability of diffusion approximation to cover wider and deeper aspects of stochastic optimization algorithms in data science.
登录
查看更多内容
DOI:
--
发表时间:
2017-04
期刊:
ArXiv
影响因子:
--
作者:
S. Mandt;M. Hoffman;D. Blei
通讯作者:
S. Mandt;M. Hoffman;D. Blei
DOI:
10.1137/130935616
发表时间:
2014-07
期刊:
SIAM J. Numer. Anal.
影响因子:
--
作者:
A. Abdulle;G. Vilmart;K. Zygalakis
通讯作者:
A. Abdulle;G. Vilmart;K. Zygalakis
DOI:
10.2139/ssrn.2954149
发表时间:
2016-11
期刊:
Econometrics: Econometric & Statistical Methods - General eJournal
影响因子:
--
作者:
Justin A. Sirignano;K. Spiliopoulos
通讯作者:
Justin A. Sirignano;K. Spiliopoulos
DOI:
--
发表时间:
2018-11
期刊:
J. Mach. Learn. Res.
影响因子:
--
作者:
Qianxiao Li;Cheng Tai;E. Weinan
通讯作者:
Qianxiao Li;Cheng Tai;E. Weinan
DOI:
10.1093/imanum/dru016
发表时间:
2013-10
期刊:
arXiv: Numerical Analysis
影响因子:
--
作者:
M. Kopec
通讯作者:
M. Kopec