Adaptive robust estimation in sparse vector model (Q820801)

This paper is a study on construction of estimators that are optimally adaptive in a minimax sense with respect to the noise level, to the form of the noise distribution, and to the sparsity. For the sparse vector model, the authors consider estimation of the target vector, of its l2-norm and of the noise variance. The noise random variables are assumed to be i.i.d.. The authors consider classes of noise distributions with polynomially and exponentially decreasing tails as well as the case of Gaussian noise. They obtained rates turn out to be different from the minimax nonadaptive rates when the triplet is known. They study estimation of sparse vector in l2-norm when the variance of noise and the distribution of random variables are both unknown. They only assume that the noise distribution belongs to a given class, which can be either a class of distributions with polynomial tails, or a class with exponential decay of the tails. They show the estimation of the l2-norm, the estimation of the variance of the noise. They also show the proofs of the upper bounds and the lower bounds.