Dual-Cloud Multi-Secret Sharing Architecture for Privacy Preserving Persistent Computation

With the prevalence of artificial intelligence, people collect data through numerous sensors and use machine learning to create models for intelligent services. However, data privacy and massive data issues are raised with the proliferation of devices. Although secret sharing can be the solution to providing privacy and performing efficient computations (than homomorphic encryption-based) in the privacy domain, sharing large amounts of data may impose a considerable storage burden on resource-restricted devices. Therefore, we aim to reduce the shares that participants need to hold and construct secure computation protocols that form privacy-preserving data utilization with multi-secret sharing. In this paper, we study the mechanism of persistent computation with multi-secret sharing for privacy protection, which avoids the requirements for participants to store extra information during secure computation. We present the mask techniques to convert secrets into protected data and upload them separately to non-colluding dual-cloud servers through multi-secret sharing. Then, we can locally evaluate the operation result shares by revealing a part of the protected data, where the true secret consistently remains secure as the other part stays in a shared state. Eventually, we outsource a dataset to the cloud, build a privacy-preserving multi-party kNN classification based on our scheme, and provide some experiments to demonstrate the feasibility and usability of storage size.