Privacy-Preserving Cloud-Based Statistical Analyses On Sensitive Categorical Data

We consider the problem of privacy-preserving cloud-based statistical computation on sensitive categorical data. Specifically, we focus on protocols to obtain the contingency matrix and the sample covariance matrix of the categorical data set. A multi-cloud is used not only to store the sensitive data but also to perform computations on them. However, the multi-cloud is semi-honest, that is, it follows the protocols but is not authorized to learn the sensitive data. Hence, the data must be stored and computed on by the multi-cloud in a privacy-preserving format, which we choose to be vertical splitting among the various clouds. We give a comparison of our proposals, based on the secure scalar product, against a benchmark protocol consisting of downloading plus local computation.