Secured Privacy Preserving Data Aggregation with Semi-honest Servers

Abstract

With the large deployment of smart devices, the collections and analysis of user data significantly benefit both industry and people’s daily life. However, it has showed a serious risk to people’s privacy in the process of the above applications. Recently, combining multiparty computation and differential privacy was a popular strategy to guarantee both computational security and output privacy in distributed data aggregation. To decrease the communication cost in traditional multiparty computation paradigm, the existing work introduces several trusted servers to undertake the main computing tasks. But we will lose the guarantee on both security and privacy when the trusted servers are vulnerable to adversaries. To address the privacy disclosure problem caused by the vulnerable servers, we provide a two-layer randomisation privacy preserved data aggregation framework with semi-honest servers (we only take their computation ability but do not trust them). Differing from the existing approach introduces differential privacy noises globally, our framework randomly adds random noises but maintains the same differential privacy guarantee. Theoretical and experimental analysis show that to achieve same security and privacy insurance, our framework provides better data utility than the existing approach.