scientific article; zbMATH DE number 3965251

From MaRDI portal
Publication:3732787





[For the entire collection see Zbl 0477.00022.] The problem of experimental design is considered under the assumption that the expectation of the observed function is an arbitrary function \(f\in L_ 2\), i.e., when there exists practically no prior information about the form of the regression model. In this case the aim of the experiment is to estimate the polynomial in a given system of orthogonal functions that is the best approximation to the function f in the \(L_ 2\) metric. The solution of this problem of nonparametric statistics is given by a randomized experimental design. In the present paper the authors compare various approaches to the approximate solution to the problem of optimal choice of a randomized design by means of a number of particular examples. Depending on the form of the desired approximation, the problem reduces either to the problem of linear, infinitely linear or nonlinear programming. Reviewer's remarks: The problem of experimental design under the above- cited assumption, as well as the concept of a randomized design exactly in the sense of the paper under review, was originally introduced in 1968 by the first author and the reviewer [Dokl. Akad. Nauk SSSR 180, 273-275 (1968; Zbl 0177.467)] and not in the first author's book The Monte Carlo method and some related problems. (Russian), second edition (1975; for a review of the 1st ed. see Zbl 0228.65004)] as claimed by the authors in the paper under review. (For references see also the above- cited book by the first author and the joint paper by the first author and the reviewer, Probability theory, mathematical statistics, theoretical cybernetics, Vol. 13 (Russian), 59-108, Akad. Nauk SSSR VINITI, Moscow (1966).)











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3732787)