Secure Computation on Additive Shares

The rapid development of cloud computing has probably benefited each of us. However, the privacy risks brought by untrustworthy cloud servers arise the attention of more and more people and legislatures. In the last two decades, plenty of works seek the way of outsourcing various specific tasks while ensuring the security of private data. The tasks to be outsourced are countless; however, the calculations involved are similar. In this paper, inspired by additive secret sharing and multiplicative secret sharing technologies, we construct a series of novel protocols that support common secure calculations on numbers (e.g., basic elementary functions) or matrices (e.g., solve eigenvectors) in arbitrary $n$ servers ($n \geq 2$). Moreover, the $n$-party protocols ensure the security of the original data even if $n-1$ servers collude. With the help of the client-aided model, the protocols can support the computations on float-pointing numbers. All proposed protocols only require constant rounds of interaction, and we give the detailed security and efficiency analysis. The convolutional neural network is utilized as a case to demonstrate the advantage. We believe that these protocols can provide a new basic tool for actual outsourced tasks.