A Data Synthesis Approach Based on Local Differential Privacy

Privacy-preserving data publishing is an important problem that has been the focus of extensive study. The state-of-the-art way for this problem is Local Differential Privacy (LDP), which can protect individual user’s privacy without relying on a trusted third party. However, the existing LDP-based works perform poorly in the setting of data publishing; even worse, some of them may incur very expensive computational overheads and also reduce greatly the utility of published data. In this paper, we propose a local differential privacy approach, LDPrivBayes, for publishing synthetic data, which improves the previous work and performs consistently better than existing approaches. We conduct extensive experiments on the different datasets. Experimental results demonstrate the effectiveness of our approach over the existing approaches.