Group-wise K-anonymity meets (ε, δ) Differentially Privacy Scheme

We studied the link between K-anonymity and differential privacy as the basis for deriving a novel method for noise estimation. Hence, we provide threefold contributions: First, we use the birthday-bound paradox for uniqueness to estimate the noise level, ε in (ε, δ) differentially privacy scheme. Second, our group-aware formulation provides resilience to a series of inference attacks by using the group privacy property in our unique group-centric formulation. Third, draw a connection between the attacker advantage, δ, and ε for univariate and multivariate cases. Finally, we demonstrate applicability in Laplacian, Gaussian, and Exponential mechanisms.