1: \begin{abstract}

2: We consider two problems of estimation in high-dimensional Gaussian models.

3: The first problem is that of estimating a linear functional of the means of

4: $n$ independent  $p$-dimensional Gaussian vectors, under the assumption that

5: most of these means are equal to zero. We show that, up to a logarithmic

6: factor, the minimax rate of estimation in squared Euclidean norm is

7: between $(s^2\wedge n) +sp$ and $(s^2\wedge np)+sp$. The estimator that attains

8: the upper bound being computationally demanding, we investigate suitable

9: versions of group thresholding estimators that are efficiently computable even

10: when the dimension and the sample size are very large. An interesting new phenomenon

11: revealed by this investigation is that the group thresholding leads to a

12: substantial improvement in the rate as compared to the element-wise thresholding.

13: Thus, the rate of the group

14: thresholding is $s^2\sqrt{p}+sp$, while the element-wise thresholding has an

15: error of order $s^2p+sp$. To the best of our knowledge, this is the first known setting in which leveraging the group structure leads to a polynomial

16: improvement in the rate.

17: 

18: The second problem studied in this work is the estimation of the common

19: $p$-dimensional mean of the inliers among $n$ independent Gaussian vectors.

20: We show that there is a strong analogy between this problem and the first one.

21: Exploiting it, we propose new strategies of robust estimation that are computationally tractable and have better rates of convergence than the other

22: computationally tractable robust (with respect to the presence of the

23: outliers in the data) estimators studied in the literature. However, this tractability comes with a loss of the minimax-rate-optimality in some regimes.

24: 

25: \end{abstract}

26: