A stochastic Remes algorithm

almost sure convergence convergence in distribution estimates of first derivatives Haar system invariance principle linear Tchebycheff approximation order of convergence rate of convergence stochastic Remes algorithm

Mathematics Subject Classification ID

Best approximation, Chebyshev systems (41A50) General nonlinear regression (62J02) Stochastic approximation (62L20) Algorithms for approximation of functions (65D15)

Review text

The linear Tchebycheff approximation \(\sum a_ kh_ k\) for a regression function f by a Haar system \(h_ 0,h_ 1,...,h_ N\) is estimated by a stochastic Remes algorithm. This procedure is related to the usual Newton-type Remes algorithm as the Kiefer-Wolfowitz method of stochastic approximation is related to the classical Newton method. Only estimates for the first derivatives of f are used. Under suitable conditions the estimates for the alternant, for the coefficients \(a_ k\) and the minimal error \(\lambda\) all converge almost surely to the true values, when the number n of iterations tends to infinity. There is also convergence in distribution, which is formulated as an invariance principle. Under usual conditions on the noise the order of convergence is \(O(n^{-1/4})\) for the alternant (even \(O(n^{-1/3})\) if f is in \(C^ 3)\), and \(O(n^{-})\) for the coefficients and the minimal error.

Recommendations

Cites work

Cited in

(1)

scientific article; zbMATH DE number 3899488 (Why is no real title available?)

This page was built for publication: A stochastic Remes algorithm

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1091725)