Verifiable computation over encrypted data via MPC-in-the-head zero-knowledge proofs

Abstract

In the current landscape of cloud-based data storage and analysis, concerns about data privacy and integrity have become more and more prevalent. Homomorphic encryption is a promising technology for preserving privacy by enabling computations on encrypted data while maintaining the confidentiality of sensitive information. However, relying solely on HE may pose challenges in ensuring the integrity of data and computation, which necessitates the verification of outsourced computations for users. In this paper, we propose a generic solution for verifiable computation over encrypted data. Our solution is based on a lattice-based approximate homomorphic encryption scheme with an MPC-in-the-Head style zero-knowledge proof system. We demonstrate that a user provided with a third party's certification of the computed function can verify the homomorphic evaluation over encrypted data. In the experiment, we provide a proof-of-concept implementation of our algorithms for privacy-preserving machine learning including regression, classification and validation. Our solution is post-quantum and can be extended to various scenarios such as privacy-preserving machine learning.