Paper 2017/979
Privacy-Preserving Ridge Regression over Distributed Data from LHE
Irene Giacomelli and Somesh Jha and Marc Joye and C. David Page and Kyonghwan Yoon
Abstract
Linear regression with 2-norm regularization (i.e., ridge regression) is an important statistical technique 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 health-care), these values represent sensitive data owned by several different parties who are unwilling to share them. In this setting, training a linear regression model becomes challenging and needs specific cryptographic solutions. This problem was elegantly addressed by Nikolaenko et al. in S&P (Oakland) 2013. They suggested a two-server system that uses linearly-homomorphic encryption (LHE) and Yao’s two-party protocol (garbled circuits). In this work, we propose a novel system that can train a ridge linear regression model using only linearly-homomorphic encryption (i.e., without using Yao’s protocol). This greatly improves the overall performance (both in computation and communications) as Yao’s protocol was the main bottleneck in the previous solution. The efficiency of the proposed system is validated both on synthetically-generated and real-world datasets.
Note: This paper is a merge of ePrint:2017/732 and ePrint:2017/707.
Metadata
- Available format(s)
- Publication info
- Preprint. MINOR revision.
- Contact author(s)
- irene giacomelli29 @ gmail com
- History
- 2018-04-14: last of 2 revisions
- 2017-10-09: received
- See all versions
- Short URL
- https://ia.cr/2017/979
- License
-
CC BY