Nonparametric estimation of the regression function in \(L^ 2\) (Q2277703)

Consider the problem of nonparametric estimation of a regression function f(.) in the interval [0,1], on the base of observations \(X^ N_ i=f(t^ N_ i)+\epsilon \xi^ N_ i\), \(i=1,...,N\), where \(\xi^ N_ i\) are independent Gaussian random variables with zero mean and unit variance, and the regressors \(t^ N_ i\in [0,1]\), \(i=1,...,N\), are nonrandomized and chosen in advance. Employing a quadratic loss function the authors construct adaptive estimators of f(.) and study their properties. Actually, it is proved that under certain conditions the risk of the constructed estimators in the principal term coincides with the risk of the best linear estimator constructed under the known regression function.