Introducing Secure XGBoost -- enabling XGBoost on encrypted data

#1

Hello XGBoost community!

In the RISE Lab at UC Berkeley, we’ve been working on Secure XGBoost, a library that augments XGBoost with layers of security, as part of the MC2 project. Excitingly, Secure XGBoost enables the training and inference of XGBoost models on encrypted data. For ease of use, we provide a Python API nearly identical to that of XGBoost, with only a few additions to integrate security.

Please check out our GitHub repo and our article on our work! We’d love any feedback from the XGBoost community.

1 Like
#2

Looks great, thanks Chester!

1 Like
#3

This is definitely a big deal for banking and life sciences. Great work!
A question: is this work related to homeomorphic encryption?

1 Like
#4

Not quite. Rather than using homomorphic encryption, our work uses specialized hardware (secure enclaves) to provide security guarantees. We made this choice because using specialized hardware (rather than homomorphic encryption) enables training and inference to run much faster.

1 Like