Secure Multi-party Key-Value Data Statistics Against Malicious Models
Jilong Zhao, Yuhao Zhang, Xiaobo Sharon Hu · Procedia Computer Science · 2024
Recently, key-value storage has become the primary paradigm for managing associative data due to its simplicity and efficiency, and is widely used in various fields. However, key-value data often contains sensitive information such as identity, health data, and transaction records. Once this data is collected, the risk of privacy breaches for users significantly increases. Differential privacy is commonly used to compute aggregated statistical data while protecting user privacy, but traditional differential privacy relies on a trusted third party, which is often not available in reality. Secure Multi-Party Computation (MPC) can analyze key value data statistics without the necessity of a central trusted authority, but it involves high computational and communication costs and risks from malicious participants. To address these issues, we design an MPC scheme for malicious models based on the SPDZ protocol. This scheme combines secure multi-party computation and differential privacy to securely compute the frequency and mean of key value data without a trusted third party, achieving differential privacy protection comparable to centralized differential privacy. Upon security analysis, it has been established that our protocol is secure within the Universal Composability (UC) framework. Once a violation of the protocol by any participant is detected, the protocol will terminate, providing resistance against disruptions from malicious participants. Comparative experiments demonstrate that this scheme performs comparably under a malicious security model as it does under a semi-honest security model, verifying its feasibility and effectiveness.