Acceleration of stochastic approximation by averaging

@article{Polyak1992AccelerationOS,
  title={Acceleration of stochastic approximation by averaging},
  author={Boris Polyak and Anatoli B. Juditsky},
  journal={Siam Journal on Control and Optimization},
  year={1992},
  volume={30},
  pages={838-855}
}
A new recursive algorithm of stochastic approximation type with the averaging of trajectories is investigated. Convergence with probability one is proved for a variety of classical optimization and identification problems. It is also demonstrated for these problems that the proposed algorithm achieves the highest possible rate of convergence. 
View via Publisher
meyn.ece.ufl.edu
1,716 Citations
Highly Influential Citations
214
Background Citations
554
Methods Citations
695
Results Citations
48

1,716 Citations

Asymptotically efficient stochastic approximation

Asymptotic efficiency of a stochastic approximation algorithm is proved under weak conditions: no growth rate restriction is imposed on the regression function; wide class of observation noises is…

On Stochastic Approximation Procedures with Averaging

Linear multidimensional stochastic approximation procedures in continuous time with martingale errors are considered. An asymptotic behavior of the estimator obtained by trajectory averaging is…

Asymptotic Expansions for the Stochastic Approximation Averaging Procedure in Continuous Time

This paper considers the statistical estimation problem of the root of a nonlinear function under observations of the nonlinear regression model in continuous time. Asymptotic expansions for…

Quasi stochastic approximation

A general foundation for "quasi-stochastic approximation" in which all of the processes under consideration are deterministic, much like quasi-Monte-Carlo for variance reduction in simulation is provided.

Asymptotic behavior of asynchronous stochastic approximation

The pathwise convergence of a distributed, asynchronous stochastic approximation (SA) scheme is analyzed. The conditions imposed on the step size and noise are the weakest in comparison with the…

Deviation inequalities for stochastic approximation by averaging

Rates of Convergence of Adaptive Step-Size of Stochastic Approximation Algorithms

A new adaptive algorithm with decreasing step-size for stochastic approximations with optimality of the rate of convergence is proposed and an averaging algorithm is used to control the step sizes.

Globally convergent stochastic optimization with optimal asymptotic distribution

  • J. Dippon
  • Mathematics, Computer Science
    Journal of Applied Probability
  • 1998
A stochastic gradient descent method is combined with a consistent auxiliary estimate to achieve global convergence of the recursion and the optimal convergence rate and variance of the asymptotic distribution.

Martingale Models of Stochastic Approximation and Their Convergence

Procedures of stochastic approximation are studied from a general theory of stochastic processes point of view. The results on convergence are obtained by the uniform methods both in the case of…

Some Limit Properties of Markov Chains Induced by Recursive Stochastic Algorithms

Recursive stochastic algorithms have gained significant attention in the recent past due to data-driven applications and can be used for solving large-scale optimizati...
...

References

SHOWING 1-10 OF 20 REFERENCES

Stochastic approximation method with gradient averaging for unconstrained problems

A stochastic approximation algorithm is studied, in which random gradient estimates are averaged by an auxiliary, filter. Convergence of the algorithm is proved under a rather general noise…

Asymptotically efficient stochastic approximation

Asymptotic efficiency of a stochastic approximation algorithm is proved under weak conditions: no growth rate restriction is imposed on the regression function; wide class of observation noises is…

Convergence analysis of smoothed stochastic gradient-type algorithm

A rigorous analysis is presented which concludes, under very mild assumptions on the data, that the smoothed stochastic gradient (SSG) algorithm converges almost everywhere.

Theory and Practice of Recursive Identification

Methods of recursive identification deal with the problem of building mathematical models of signals and systems on-line, at the same time as data is being collected. Such methods, which are also k…

Adaptive filtering prediction and control

This unified survey focuses on linear discrete-time systems and explores the natural extensions to nonlinear systems and summarizes the theoretical and practical aspects of a large class of adaptive algorithms.

Stochastic Approximation: A Recursive Method for Solving Regression Problems1

Stochastic Approximation and Recursive Estimation

An extrusion head for an extrudable material, such as a food product dough, which is to be formed into a special shape, such as a pretzel configuration. The head includes a die and associated parts…

Stochastic Approximation and Recursive Estimation

Stochastic Programming Methods

An extension of the Robbins-Monro procedure