Differentially private binary- and matrix-valued data query

Differential privacy has been widely adopted to release continuous- and scalar-valued information on a database without compromising the of individual data records in it. The problem querying binary- matrix-valued differentially private manner rarely studied. However, are ubiquitous real-world applications, whose concerns may arise under variety circumstances. In this paper, we devise an exclusive or (XOR) mechanism that perturbs query result by conducting XOR operation with calibrated noises attributed Bernoulli distribution. We first rigorously analyze utility guarantee proposed mechanism. Then, generate parameters distribution, develop heuristic approach minimize expected square error rate ϵ -differential constraint. Additionally, address intractability calculating probability density function (PDF) distribution efficiently samples from it, adapt Exact Hamiltonian Monte Carlo based sampling scheme. Finally, experimentally demonstrate efficacy considering binary classification social network analysis, all manner. Experiment results show notably outperforms other state-of-the-art methods terms (such as accuracy F 1 score), even achieves comparable non-private mechanisms.