Bridging the gap between constant step size stochastic gradient descent and Markov chains (Q2196224): Difference between revisions

From MaRDI portal
Import240304020342 (talk | contribs)
Set profile property.
ReferenceBot (talk | contribs)
Changed an Item
 
(One intermediate revision by one other user not shown)
Property / arXiv ID
 
Property / arXiv ID: 1707.06386 / rank
 
Normal rank
Property / cites work
 
Property / cites work: High Order Numerical Approximation of the Invariant Measure of Ergodic SDEs / rank
 
Normal rank
Property / cites work
 
Property / cites work: On a Perturbation Approach for the Analysis of Stochastic Tracking Algorithms / rank
 
Normal rank
Property / cites work
 
Property / cites work: Adaptivity of averaged stochastic gradient descent to local strong convexity for logistic regression / rank
 
Normal rank
Property / cites work
 
Property / cites work: A Dynamical System Approach to Stochastic Approximations / rank
 
Normal rank
Property / cites work
 
Property / cites work: Q3997575 / rank
 
Normal rank
Property / cites work
 
Property / cites work: Q3151174 / rank
 
Normal rank
Property / cites work
 
Property / cites work: About the multidimensional competitive learning vector quantization algorithm with constant gain / rank
 
Normal rank
Property / cites work
 
Property / cites work: Statistical inference for model parameters in stochastic gradient descent / rank
 
Normal rank
Property / cites work
 
Property / cites work: Theoretical Guarantees for Approximate Sampling from Smooth and Log-Concave Densities / rank
 
Normal rank
Property / cites work
 
Property / cites work: Nonparametric stochastic approximation with large step-sizes / rank
 
Normal rank
Property / cites work
 
Property / cites work: Harder, Better, Faster, Stronger Convergence Rates for Least-Squares Regression / rank
 
Normal rank
Property / cites work
 
Property / cites work: Nonasymptotic convergence analysis for the unadjusted Langevin algorithm / rank
 
Normal rank
Property / cites work
 
Property / cites work: Convergence of stochastic algorithms: from the Kushner–Clark theorem to the Lyapounov functional method / rank
 
Normal rank
Property / cites work
 
Property / cites work: Asymptotic Behavior of a Markovian Stochastic Algorithm with Constant Step / rank
 
Normal rank
Property / cites work
 
Property / cites work: Q4225410 / rank
 
Normal rank
Property / cites work
 
Property / cites work: A Liapounov bound for solutions of the Poisson equation / rank
 
Normal rank
Property / cites work
 
Property / cites work: Ordinary Differential Equations / rank
 
Normal rank
Property / cites work
 
Property / cites work: Q4558562 / rank
 
Normal rank
Property / cites work
 
Property / cites work: Stochastic approximation methods for constrained and unconstrained systems / rank
 
Normal rank
Property / cites work
 
Property / cites work: An optimal method for stochastic composite optimization / rank
 
Normal rank
Property / cites work
 
Property / cites work: Analysis of recursive stochastic algorithms / rank
 
Normal rank
Property / cites work
 
Property / cites work: Q4003497 / rank
 
Normal rank
Property / cites work
 
Property / cites work: Q4637063 / rank
 
Normal rank
Property / cites work
 
Property / cites work: Ergodicity for SDEs and approximations: locally Lipschitz vector fields and degenerate noise. / rank
 
Normal rank
Property / cites work
 
Property / cites work: Applications of a Kushner and Clark lemma to general classes of stochastic algorithms / rank
 
Normal rank
Property / cites work
 
Property / cites work: Théorèmes de convergence presque sure pour une classe d'algorithmes stochastiques à pas decroissant / rank
 
Normal rank
Property / cites work
 
Property / cites work: Markov Chains and Stochastic Stability / rank
 
Normal rank
Property / cites work
 
Property / cites work: On recursive estimation for time varying autoregressive processes / rank
 
Normal rank
Property / cites work
 
Property / cites work: Q2752037 / rank
 
Normal rank
Property / cites work
 
Property / cites work: Stochastic gradient descent, weighted sampling, and the randomized Kaczmarz algorithm / rank
 
Normal rank
Property / cites work
 
Property / cites work: Robust Stochastic Approximation Approach to Stochastic Programming / rank
 
Normal rank
Property / cites work
 
Property / cites work: Q3967358 / rank
 
Normal rank
Property / cites work
 
Property / cites work: Confidence level solutions for stochastic programming / rank
 
Normal rank
Property / cites work
 
Property / cites work: Stochastic Minimization with Constant Step-Size: Asymptotic Laws / rank
 
Normal rank
Property / cites work
 
Property / cites work: Acceleration of Stochastic Approximation by Averaging / rank
 
Normal rank
Property / cites work
 
Property / cites work: A remark on the stability of the l.m.s. tracking algorithm / rank
 
Normal rank
Property / cites work
 
Property / cites work: A Stochastic Approximation Method / rank
 
Normal rank
Property / cites work
 
Property / cites work: Pegasos: primal estimated sub-gradient solver for SVM / rank
 
Normal rank
Property / cites work
 
Property / cites work: Q4779819 / rank
 
Normal rank
Property / cites work
 
Property / cites work: Asymptotic bias of stochastic gradient search / rank
 
Normal rank
Property / cites work
 
Property / cites work: Expansion of the global error for numerical schemes solving stochastic differential equations / rank
 
Normal rank
Property / cites work
 
Property / cites work: Optimal Transport / rank
 
Normal rank
Property / cites work
 
Property / cites work: Co-Coercivity and Its Role in the Convergence of Iterative Schemes for Solving Variational Inequalities / rank
 
Normal rank

Latest revision as of 10:17, 23 July 2024

scientific article
Language Label Description Also known as
English
Bridging the gap between constant step size stochastic gradient descent and Markov chains
scientific article

    Statements

    Bridging the gap between constant step size stochastic gradient descent and Markov chains (English)
    0 references
    0 references
    0 references
    0 references
    28 August 2020
    0 references
    The paper deals with the minimization algorithm of a strongly convex objective function given access to unbiased estimates of its gradient through \textit{Stochastic Gradient Descent} (SGD), aka Robbins-Monro algorithm [\textit{H. Robbins} and \textit{S. Monro}, Ann. Math. Stat. 22, 400--407 (1951; Zbl 0054.05901)], with constant step size. As the detailed analysis was only performed for quadratic functions, the authors provide an explicit asymptotic expansion of the moments of the averaged SGD iterates that outlines the dependence on initial conditions, the effect of noise and the step size, as well as the lack of convergence in the general (nonquadratic) case. This analysis is based on tools from Markov chain theory into the analysis of stochastic gradient. It is observed that Richardson-Romberg extrapolation [\textit{G. Pagès}, Monte Carlo Methods Appl. 13, No. 1, 37--70 (2007; Zbl 1119.65004); \textit{N. Frikha} and \textit{L. Huang}, Stochastic Processes Appl. 125, No. 11, 4066--4101 (2015; Zbl 1336.60137)] may be used to get closer to the global optimum. Empirical improvements of the new extrapolation scheme is shown. This methodological problem is of interest in different practical tasks appearing in large-scale machine learning, optimization and stochastic approximation.
    0 references
    stochastic gradient descent
    0 references
    Markov chains
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references

    Identifiers

    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references