Secure computation of inner product of vectors with distributed entries and its applications to SVM

Abstract

Nowadays organizations and individuals outsource computation and storage to cloud. This poses a threat to the privacy of users. Different users encrypt their private data with (possibly) different keys to prevent any kind of outside attack on their privacy. In this outsourced model of computation where the data owners have already encrypted and uploaded private data, to enable the users for collaborative data mining a scheme is needed that can process encrypted data under multiple keys. Privacy preserving inner product computation is an essential tool on which many data mining algorithms are based. Several papers address the problem of outsourced privacy preserving inner product computation but none of them deals with the scenario when the entire database is arbitrarily partitioned among the users. We propose two outsourced privacy preserving protocols for computation of inner product of vectors when the underlying database is arbitrarily partitioned. We provide an SVM training model that preserves the privacy of the user’s data-vectors. Our scheme is based on an integer vector encryption scheme.