1: \begin{abstract}

2:   We design an $(\diffp, \delta)$-differentially private algorithm

3:   to estimate the mean of a $d$-variate distribution, with unknown covariance

4:   $\Sigma$, that is adaptive to $\Sigma$. To within polylogarithmic factors,

5:   the estimator achieves optimal rates of convergence with

6:   respect to the induced Mahalanobis norm $\norm{\cdot}_\Sigma$,

7:   takes time $\wt{O}(n d^2)$ to compute, has near linear

8:   sample complexity for sub-Gaussian distributions, allows

9:   $\Sigma$ to be degenerate or low rank, and adaptively

10:   extends beyond sub-Gaussianity. Prior to this work, other methods

11:   required exponential computation time or the superlinear

12:   scaling $n = \Omega(d^{3/2})$ to achieve non-trivial error

13:   with respect to the norm $\norm{\cdot}_\Sigma$.

14: \end{abstract}

15: