Improvement on a privacy-preserving outsourced classification protocol over encrypted data

Abstract

In outsourced classification services, classifier owners outsource their classifiers to remote servers due to resource constraints, and users can request classification services from this server. What attracts us is that the users’ query data, classification results, and classifier privacy are all well protected during classification. However, we introduce a threat model that makes it easy for adversaries to attack. Thus, to ensure its security, this model should be modified. In addition, considering the low efficiency of Paillier cryptosystem, the classification phase is accompanied by problems of low computational efficiency and large occupied bandwidth consumption. In this paper, we use a substitutive OU cryptosystem, which effectively saves computational and communication costs. Moreover, experimental results show that the improvement enhances the efficiency of the scheme and reduces the bandwidth consumption.