Secure Collaborative Training and Inference for XGBoost

TL;DR

Secure XGBoost is a privacy-preserving system enabling multiparty training and inference of XGBoost models, utilizing hardware enclaves and novel data-oblivious algorithms to prevent side-channel attacks.

Contribution

It introduces a secure system for collaborative XGBoost training and inference that combines hardware enclaves with data-oblivious algorithms for enhanced security.

Findings

Protects data privacy during collaborative training

Prevents access pattern side-channel attacks

Enables secure multiparty inference

Abstract

In recent years, gradient boosted decision tree learning has proven to be an effective method of training robust models. Moreover, collaborative learning among multiple parties has the potential to greatly benefit all parties involved, but organizations have also encountered obstacles in sharing sensitive data due to business, regulatory, and liability concerns. We propose Secure XGBoost, a privacy-preserving system that enables multiparty training and inference of XGBoost models. Secure XGBoost protects the privacy of each party's data as well as the integrity of the computation with the help of hardware enclaves. Crucially, Secure XGBoost augments the security of the enclaves using novel data-oblivious algorithms that prevent access side-channel attacks on enclaves induced via access pattern leakage.