Towards Vertical Privacy-Preserving Symbolic Regression via Secure Multiparty Computation

TL;DR

This paper introduces a secure multiparty computation approach for vertical privacy-preserving symbolic regression, enabling multiple parties to collaboratively build models without exposing their private data, thus addressing privacy concerns in distributed settings.

Contribution

It is the first to apply secure multiparty computation to vertical symbolic regression, extending privacy-preserving techniques beyond horizontal data partitioning.

Findings

Achieves comparable accuracy to centralized models

Ensures data privacy during collaborative modeling

Demonstrates feasibility with preliminary experiments

Abstract

Symbolic Regression is a powerful data-driven technique that searches for mathematical expressions that explain the relationship between input variables and a target of interest. Due to its efficiency and flexibility, Genetic Programming can be seen as the standard search technique for Symbolic Regression. However, the conventional Genetic Programming algorithm requires storing all data in a central location, which is not always feasible due to growing concerns about data privacy and security. While privacy-preserving research has advanced recently and might offer a solution to this problem, their application to Symbolic Regression remains largely unexplored. Furthermore, the existing work only focuses on the horizontally partitioned setting, whereas the vertically partitioned setting, another popular scenario, has yet to be investigated. Herein, we propose an approach that employs a privacy-preserving technique called Secure Multiparty Computation to enable parties to jointly build Symbolic Regression models in the vertical scenario without revealing private data. Preliminary experimental results indicate that our proposed method delivers comparable performance to the centralized solution while safeguarding data privacy.