Paper 2017/335

Privacy-Preserving Linear Regression on Distributed Data

Irene Giacomelli, Somesh Jha, and C. David Page

Abstract

Linear regression is an important statistical tool that models the relationship between some explanatory values and an outcome value using a linear function. In many current applications (e.g. predictive modelling in personalized healthcare), these values represent sensitive data owned by several different parties that are unwilling to share them. In this setting, training a linear regression model becomes challenging and needs specific cryptographic solutions. In this work, we propose a new system that can train two different variants of linear regression (i.e. ridge regression and lasso regression) on a dataset obtained by merging a finite number of private datasets. Our system assures that no extra information on a single private dataset is revealed to the entities performing the learning algorithm. Moreover, our solution is based on efficient cryptographic tools (e.g. Paillier’s scheme and pseudorandom generator).