Privacy-Preserving and Verifiable Outsourcing Linear Inference Computing Framework

Jiao Liu,Xinghua Li,Ximeng Liu,Jiawei Tang,Yunwei Wang,Qiuyun Tong,Jianfeng Ma

DOI: https://doi.org/10.1109/tsc.2023.3332933

IF: 11.019

2023-01-01

IEEE Transactions on Services Computing

Abstract:In machine learning (ML), the massive data processing and dense computations based on matrices make outsourced inference computation a growing trend. The unreliability of cloud platforms makes privacy protection and inference correctness increasingly important in outsourced computations. Unfortunately, current works cannot provide an effective verification mechanism and privacy protection for outsourcing linear computing simultaneously. To address the issue, in the service architectures with malicious behaviors (such as curiosity, dishonesty, and collusion), we propose privacy-preserving and verifiable outsourcing inference computing (PPVLC) for the most fundamental linear computations in ML. PPVLC uses secret sharing and blinding techniques to protect privacy and achieve secure computation of matrix linear computation. Meanwhile, bilinear mapping based on matrix digest is utilized to verify computation correctness, ensuring the trustworthiness of the service. Security analysis and experiments demonstrate the reliability of our scheme and service efficiency.