Construction of Differentially Private Summaries over Fully Homomorphic Encryption

TL;DR

This paper introduces DP-summary, a method for creating differentially private data summaries over fully homomorphic encryption, improving security and accuracy in cloud query processing without decrypting data.

Contribution

It proposes a novel approach combining homomorphic encryption with the DAWA algorithm to generate private data summaries without decryption, reducing noise and errors.

Findings

Successfully summarized eight 16-bit data in 12.5 hours using TFHE.

No accuracy degradation observed with the proposed method.

Enhanced data security by avoiding data decryption during summarization.

Abstract

Cloud computing has garnered attention as a platform of query processing systems. However, data privacy leakage is a critical problem. Chowdhury et al. proposed Crypt(epsilon), which executes differential privacy (DP) over encrypted data on two non-colluding semi-honest servers. Further, the DP index proposed by these authors summarizes a dataset to prevent information leakage while improving the performance. However, two problems persist: 1) the original data are decrypted to apply sorting via a garbled circuit, and 2) the added noise becomes large because the sorted data are partitioned with equal width, regardless of the data distribution. To solve these problems, we propose a new method called DP-summary that summarizes a dataset into differentially private data over a homomorphic encryption without decryption, thereby enhancing data security. Furthermore, our scheme adopts Li et al.'s data-aware and workload-aware (DAWA) algorithm for the encrypted data, thereby minimizing the noise caused by DP and reducing the errors of query responses. An experimental evaluation using torus fully homomorphic encryption (TFHE), a bit-wise fully homomorphic encryption library, confirms the applicability of the proposed method, which summarized eight 16-bit data in 12.5 h. We also confirmed that there was no accuracy degradation even after adopting TFHE along with the DAWA algorithm.