Authors : Ion Badoi; Mihail-Iulian Plesa

Volume/Issue : Volume 6 - 2021, Issue 4 - April

Google Scholar : http://bitly.ws/9nMw

Scribd : https://bit.ly/2Rs2sUz

Abstract : Because cloud-based services are becoming more and more used in the field of machine learning the issue of data confidentiality arises. In this paper, we address the problem of privacy-preserving spam classification. One of the most used algorithms for solving this problem is logistic regression. In this work, we suppose that a remote service has a pre-trained logistic regression model about which it does not want to leak any information. On the other hand, a user wants to use the pre-trained model without revealing anything about his mail. To solve this problem, we propose a system that uses somewhat homomorphic encryption to encrypt the user data and at the same time allows the service to apply the model without finding out any information about the user mail. The main contribution of this paper is a practical tutorial on how to implement the inference of a logistic regression model over encrypted data using the EVA compiler

Keywords : Logistic Regression, Homomorphic Encryption, Privacy-Preserving, Spam Classification