1: \begin{abstract}

2: In this work, we study local minimax convergence estimation rates subject to $\epsilon$-differential privacy. Unlike worst-case rates, which may be conservative, algorithms that are locally minimax optimal must adapt to \emph{easy} instances of the problem.  We construct locally minimax differentially private estimators for one-parameter exponential families and estimating the tail rate of a distribution.  In these cases, we show that optimal algorithms for simple hypothesis testing, namely the recent optimal private testers of \citet{Canonne:2019}, directly inform the design of locally minimax estimation algorithms.

3: \end{abstract}

4: