Online Efficient Secure Logistic Regression based on Function Secret Sharing

TL;DR

This paper introduces an online efficient privacy-preserving logistic regression protocol using Function Secret Sharing, significantly reducing communication overhead and enabling secure large-scale data training.

Contribution

It presents a novel online protocol for secure logistic regression based on FSS, improving efficiency and privacy in multi-party settings with minimal online communication.

Findings

Reduces online communication overhead compared to traditional methods

Provides accurate, MPC-friendly sigmoid function approximations

Demonstrates efficiency and effectiveness through theoretical and experimental analysis

Abstract

Logistic regression is an algorithm widely used for binary classification in various real-world applications such as fraud detection, medical diagnosis, and recommendation systems. However, training a logistic regression model with data from different parties raises privacy concerns. Secure Multi-Party Computation (MPC) is a cryptographic tool that allows multiple parties to train a logistic regression model jointly without compromising privacy. The efficiency of the online training phase becomes crucial when dealing with large-scale data in practice. In this paper, we propose an online efficient protocol for privacy-preserving logistic regression based on Function Secret Sharing (FSS). Our protocols are designed in the two non-colluding servers setting and assume the existence of a third-party dealer who only poses correlated randomness to the computing parties. During the online phase, two servers jointly train a logistic regression model on their private data by utilizing pre-generated correlated randomness. Furthermore, we propose accurate and MPC-friendly alternatives to the sigmoid function and encapsulate the logistic regression training process into a function secret sharing gate. The online communication overhead significantly decreases compared with the traditional secure logistic regression training based on secret sharing. We provide both theoretical and experimental analyses to demonstrate the efficiency and effectiveness of our method.