1: \begin{abstract}%

2:   \vspace{-.3em}

3:   We consider the problem of minimizing a convex risk with stochastic subgradients guaranteeing $\epsilon$-locally differentially private ($\epsilon$-LDP).

4:   While it has been shown that stochastic optimization is possible with $\epsilon$-LDP via the standard SGD~\citep{SongCS13}, its convergence rate largely depends on the learning rate, which must be tuned via repeated runs.

5:   Further, tuning is detrimental to privacy loss since it significantly increases the number of gradient requests.

6:   In this work, we propose BANCO (Betting Algorithm for Noisy COins), the first $\epsilon$-LDP SGD algorithm that essentially matches the convergence rate of the tuned SGD without any learning rate parameter, reducing privacy loss and saving privacy budget.

7: \end{abstract}