No label defined (Q4940648)

\textit{M.S. Pinsker} [Probl. Inf. Transm. 16, No. 2, 120-133 (1980; Zbl 0452.94003)] initiated the investigation of minimax estimation procedures for the filtration problem in Gaussian noise. The statistical estimation problem considered by Pinsker is to recover the signal \(f(t)\) based on the observation of \(X_{\varepsilon}(t)\), \(0\leq t\leq T\), where \[ dX_{\varepsilon}(t)=f(t)dt +\varepsilon dW(t), \quad 0\leq t\leq T, \] \(f(t)\) is an unknown function which is assumed to belong to a known set \(\Theta\subset L_2([0,T]),\) and \(W(t)\) is a standard Wiener process. It was shown that the quadratic minimax risk over the ellipsoid \(\Theta\) coincides asymptotically with the minimax risk within the class of linear estimators. Different approaches to the minimax filtering problem were proposed by \textit{J. Franke} [J. Time Ser. Anal. 5, 227---244 (1984; Zbl 0576.62090)] and by {M. Moklyachuk} [Teor. Sluch. Protsess. 3(19), No. 3-4, 330-338 (1997; Zbl 0928.62082); Ukr. Math. J. 43, No. 1, 75-81 (1991); translation from Ukr. Math. Zh. 43, No. 1, 92-99 (1991; Zbl 0727.62091)]. More details may be found in the survey article by \textit{S.A. Kassam} and \textit{H.V. Poor} [Proc. IEEE 73, No. 3, 433-481 (1985; 569.62084)]. Problems of minimax interpolation and filtering similar to those considered by Pinsker were investigated by \textit{O.M. Kurkin, Yu.B. Korobochkin} and \textit{S.A. Shatalov} [Minimaksnaya Obrabotka Informatsii. Moskva, Energoatomizdat, 214 p. (1990)]. In Chapter 2 of this book a generalization of the white noise model to the ``coloured'' noise case is considered. The main novelty in this chapter is the derivation of the exact asymptotic behaviour of the second order term of the minimax risk. Studying the second order asymptotics enables one to improve the accuracy of the estimation and to make the structure of the optimal estimator more precise when, for example, there is a class of first order optimal estimators, but only a particular choice of the bandwidth for the kernel method makes an estimator second-order optimal. To illustrate the application of the main results, a list of examples is given which includes the well known Sobolev and analytic classes for different levels of coloured noise. The results of this chapter are also presented in \textit{E.N. Belitser} and \textit{B.Y. Levit} [Math. Methods Stat. 4, No. 3, 259-273 (1995; Zbl 0836.62070)]. In the next chapter the classical additive regression model is considered. It is supposed that the unknown regression curve lies in an ellipsoid from \(L_2([0,1])\). A distinguishing feature of the main result is that it covers a rather general class of ellipsoids. The periodic Sobolev class and the class of periodic analytic functions can be described as certain examples of such ellipsoids. The exact asymptotic behaviour of the minimax risk is described and the asymptotically minimax estimator is derived. The estimator proves to be a windowed Fourier projection estimator. The main result of this chapter can be found in \textit{E.N. Belitser} and \textit{B.Y. Levit} [Statistics 28, No. 2, 105-122 (1996; Zbl 0880.62040)]. In the last chapter the author deals with one of the basic problems in the theory of nonparametric estimation, the density estimation problem, when the observation model is complicated by the presence of censoring. The majority of the proposed density estimators in the random censoring model is based on the well-known Kaplan-Meier estimator. The author explores the local minimax risk as a measure of the quality of an estimator which yields more exact results. He describes the asymptotic behaviour of the local minimax risk and proposes an efficient estimator -- an integral of a properly chosen kernel with respect to the Kaplan-Meier estimator. A wide class of kernels on which the estimator can be based is proposed. See also \textit{E.N. Belitser} [Bernoulli 4, No.4, 519-543 (1998)].