MammothDP protects health data with private boosted decision trees

MammothDP: Private Boosted Decision Trees Ready for Trusted Hardware and Leakage-free Hyperparameters

Cryptography and SecurityDatabasesMachine Learning

Summary

Handling sensitive health data safely is a big challenge, especially when analyzing it and sharing study results. The authors developed MammothDP, a system that uses special secure computer parts and smart mathematical techniques to protect this data during processing and when releasing study information. It improves privacy by stopping the learning process early without extra privacy cost and defends against sneaky timing attacks. This makes it easier to run machine learning on health data while keeping it private and secure.

What this means in practice

  • For hospital data teams: Use MammothDP to train privacy-preserving predictive models on patient data within trusted hardware environments.
  • For financial data analysts: Apply MammothDP to develop confidential risk models while protecting sensitive tabular data under potential server threats.

Authors

Moritz Kirschte, Thorsten Peinemann, Kari Kostiainen, Jan Wichelmann, Thomas Eisenbarth, Esfandiar Mohammadi

Abstract

Data spaces like the European Health Data Space (EHDS) require the protection of sensitive data in studies during the data's processing and the study's release alike. In this work, we investigate how to implement this mandate using established trusted execution environments (TEEs) and timing side-channel hardening to protect the processing even under a compromised server, and the state-of-the-art differential privacy (DP) notion to protect the release of studies on tabular data. This combined setup could seem to complete the picture, but in fact raises unexpected security challenges that we mitigate by introducing MammothDP: a significant milestone towards a holistic machine learning system based on the recent gradient boosted decision tree learner S-BDT that is provably DP if used within a privacy filter, timing side-channel hardened, and deployed on a TEE. To achieve a strong privacy-utility tradeoff, we introduce two novel components: an early stopping criterion that adaptively selects the number of trees without any privacy overhead, and an improved rollback protection that is no longer vulnerable under timing side-channel attacks.