Differentially private partition selection

Abstract Many data analysis operations can be expressed as a GROUP BY query on an unbounded set of partitions, followed by per-partition aggregation. To make such differentially private, adding noise to each aggregation is not enough: we also need sure that the partitions released private. This problem new, and it was recently formally introduced private union [14]. In this work, continue area study, focus common setting where user associated with single partition. setting, propose simple, optimal mechanism maximizes number partitions. We discuss implementation considerations, well possible extension approach contributes fixed, small