Securely outsourcing machine learning with multiple users

Abstract

In recent years, machine learning has been widely used in data analysis for predicting models, such
as face/pattern recognition, image processing, simultaneous interpretation and speech recognition.
However, these massive data are sensitive, which raises privacy concerns. Therefore, to protect the
data privacy, in this paper, we design a scheme for securely training machine learning model on the
jointed data that provided from different sources. Our scheme falls in the two-server-aided model
and allows one server to conduct most of computations, and another server to provides auxiliary
computation. We prove the security of our scheme in the semi-honest model.