On a discrete stochastic approximation and its application to data analysis

From MaRDI portal





The stochastic approximation, introduced by {\textit{H. Robbins} and \textit{S. Monro} [Ann. Math. Stat. 22, 400-407 (1951; Zbl 0054.05901)], is a method to search for the unique solution of the functional equation \(M(x) =\alpha\) under noisy observation. The method indicates how to construct an approximation sequence \({X_n}\) of random variables that incorporates the observed value sequence \(Y_n = M(X_n) + Z_n\), disturbed by the noise \(Z_n\) (supposed to be a sequence of independent and centered random variables). The article focuses on a discrete version of the classical approximation method designed to be suitable for data analysis. The authors modify the stochastic approximation method such that the observed sequence values to be computed with a new discretization function defined by the prefixed points of observation at each step \(n\) of the observed value sequence \(Y_n\). For the procedure obtained called discrete stochastic approximation, the convergence property is shown to hold, and its mean square error is estimated. The proposed discrete stochastic approximation is proved to be useful to several data analysis applications, viz. (a) the problem of starting point detection, (b) the change point problem, and to (c) the parameter estimation of stochastic differential equations.}











This page was built for publication: On a discrete stochastic approximation and its application to data analysis

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4457880)