Privacy-preserving distributed statistical computation to a semi-honest multi-cloud

We present the problem of privacy-preserving distributed statistical computing (PPDSC) in which one party vertically splits a data set among a set of honest-butcurious clouds and wishes to use the clouds´ processing power to perform statistical computation on the overall data set. The cornerstone is to compute covariances and, more specifically, scalar products. Existing protocols for computing scalar products on split data are identified and compared, and new variants specifically designed for PPDSC are presented that improve privacy and performance.