Poster: gbdt-rs: Fast and Trustworthy Gradient Boosting Decision Tree

For data analysis services, the capability of preserving user privacy has become an eagerly demanded trait. However, most existing approaches are either unsound to close some major information leakage chanels or prohibitively expensive for practical deployment. We propose gbdt-rs , a secure and fast implementation of the gradient boosting decision tree algorithm, which is widely used in data mining and machine learning tasks. gbdt-rs is designed to fit traditional computing platforms as well as hardware-enforced trusted environments, i.e., Intel SGX secure enclaves. gbdt-rs is coded in the Rust programming language to exterminate memory errors, while its performance is barely sacrificed due to sophisticatedly optimized memory access patterns. Our experiments show that gbdt-rs can provide up to 10x speedup on inference tasks compared to XGBoost, a reputed C++ implementation of the same algorithm. In addition, gbdt-rs is highly auditable due to its relatively small code base.