scientific article; zbMATH DE number 3870402

From MaRDI portal
Publication:3336499





The authors consider the regression model: \(y_ j=y(x_ j)=\eta (x_ j)+\epsilon (x_ j)\), \(j=1,...,N\), where \(y_ j\) is the observation if \(x=x_ j\), \(\eta\) (x), \(x\in X\subset R_ k\), is the regression function belonging to some infinite-dimensional space of functions \({\mathcal F}\), and \(\epsilon\) (x) is the error with \(E\epsilon (x_ i)=0\), \(E\epsilon (x_ i)\epsilon (x_ j)=0\), \(E\epsilon^ 2(x_ i)=\sigma^ 2\), \(i\neq j=1,...,N.\) Under the assumptions that there exists a sequence of nondecreasing finite-dimensional linear subspaces generating \({\mathcal F}\) an estimator of \(\eta\) is constructed by means of a sequence of least squares estimators. Some properties of the squared risk of this estimator are studied. Asymptotically optimal designs are constructed when \({\mathcal F}\) is the space of continuous functions on \([0,1]^ k\) with period 1 and with special restrictions on their Fourier coefficients.











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 Q3336499)