We study the construction of confidence intervals under Huber's contamination model. When the contamination proportion is unknown, we characterize the necessary adaptation cost of the problem. In particular, for Gaussian location model, the optimal length of an adaptive confidence interval is proved to be exponentially wider than that of a non-adaptive one. Results for general location models will be discussed. In addition, we also consider the same problem in a network setting for an Erdos-Renyi graph with node contamination. It will be shown that the hardness of the adaptive confidence interval construction is implied by the detection threshold between Erdos-Renyi model and stochastic block model.[-]

We address the problem of estimating the distribution of presumed i.i.d. observations within the framework of Bayesian statistics. We propose a new posterior distribution that shares some similarities with the classical Bayesian one. In particular, when the statistical model is exact, we show that this new posterior distribution concentrates its mass around the target distribution, just as the classical Bayes posterior would do. However, unlike the Bayes posterior, we prove that these concentration properties remain stable when the equidistribution assumption is violated or when the data are i.i.d. with a distribution that does not belong to our model but only lies close enough to it. The results we obtain are non-asymptotic and involve explicit numerical constants.[-]