Secure XGBoost is a privacy-preserving gradient boosting library based off the popular XGBoost project. In addition to offering the same efficiency, flexibility, and portability that vanilla XGBoost provides, Secure XGBoost enables privacy-preserving model training and inference by leveraging hardware enclaves and data-oblivious algorithms.

In a nutshell, data owners can use Secure XGBoost to train a model on a remote server without revealing their data contents to the remote server. Furthermore, multiple data owners can use the library to collaboratively train a model on their collective data, without revealing their individual data to each other.

This project is currently under development as part of the broader mc² effort (i.e., Multiparty Collaboration and Coopetition) by the UC Berkeley RISE Lab.

Please feel free to reach out to us if you would like use Secure XGBoost for your applications. We also welcome contributions to our work!

To get started with the library, please refer to the documentation.