Locally differentially private high-dimensional data synthesis

In local differential privacy (LDP), a challenging problem is the ability to generate high-dimensional data while efficiently capturing correlation between attributes in dataset. Existing solutions for low-dimensional synthesis, which partition budget among all attributes, cease be effective scenarios due large-scale noise and communication cost caused by high dimension. fact, characteristics not only bring challenges but also make it possible apply some technologies break this bottleneck. This paper presents SamPrivSyn synthesis under LDP, composed of marginal sampling module generation module. The used sample from original obtain two-way marginals. process based on mutual information, updated iteratively retain, as much possible, attributes. reconstruct synthetic dataset sampled Furthermore, study conducted comparison experiments real-world datasets demonstrate effectiveness efficiency proposed method, with results proving that can protect retain information