Improving the Privacy Loss Under User-Level DP Composition for Fixed Estimation Error

arXiv:2405.06261v1 Announce Type: new
Abstract: This paper considers the private release of statistics of several disjoint subsets of a datasets, under user-level $\epsilon$-differential privacy (DP). In particular, we consider the user-level differentially private release of sample means and variances of speed values in several grids in a city, in a potentially sequential manner. Traditional analysis of the privacy loss due to the sequential composition of queries necessitates a privacy loss degradation by a factor that equals the total number of …

arxiv city cs.cr cs.it datasets differential privacy epsilon error loss math.it privacy private release sample speed statistics under