Differential Privacy Framework using Secure Computing on Untrusted Servers

Smart cities use cutting-edge technologies to monitor the real world and improve various aspects of human life, such as traffic, agriculture, electric power grid, and medical care. However, protecting users' privacy is crucial in smart-city applications for service coordination. Any leakage of users' sensitive information may prevent users from sharing their data, and thereby hinder the progress of smart cities. Differential privacy (DP) is one of the approaches that protects sensitive information in a dataset, and strongly resists background knowledge attacks. This study proposes a central differential privacy scheme using secure computing methods. After receiving the ciphertext of the data, the central data curator in this system generates a perturbed dataset in the encrypted domain. The unmodified data of each user is not exposed to the data curator, therefore the data users do not need a trusted data curator. Moreover, compared with the existing models in which significant amount of noise is locally added before sending the data to the data curator, the proposed scheme adds considerably less noise while maintaining the same privacy requirement, resulting in enhanced data utility. Model evaluation demonstrates that our system effectively protects the privacy of individuals and provides accurate data for statistical analysis.