1: \begin{abstract}

2: We study empirical and hierarchical Bayes approaches to the problem of estimating an infinite-dimensional parameter

3: in mildly ill-posed inverse problems. We consider a class of prior distributions indexed by a hyperparameter

4: that quantifies regularity. We prove that both methods we consider succeed in automatically selecting

5: this parameter optimally, resulting in optimal convergence rates for truths with  Sobolev or analytic ``smoothness'',

6: without using knowledge about this regularity. Both methods are illustrated by simulation examples.

7: \end{abstract}