MammothDP hardens private boosted trees for trusted hardware
Original title: MammothDP: Private Boosted Decision Trees Ready for Trusted Hardware and Leakage-free Hyperparameters
It adds privacy-free early stopping and rollback protection intended to resist timing side channels on TEEs.
MammothDP addresses mandates such as the European Health Data Space to protect sensitive tabular data during processing and when study results are released. It combines trusted execution environments, timing side-channel hardening, and differential privacy around the S-BDT gradient-boosted decision tree learner. The authors say S-BDT is provably differentially private inside a privacy filter. New components choose the number of trees without extra privacy cost and revise rollback protection so it is no longer vulnerable to timing attacks.
- The learner is the gradient boosted decision tree system S-BDT.
- Processing is intended to remain protected on a compromised server inside a TEE.
- Early stopping selects tree count without consuming additional privacy budget.
- Revised rollback protection targets timing side-channel exposure.
- Differential privacy is applied to released studies on tabular data.
- Organizations
- European Health Data Space
From the original article
1 min read at arxiv.orgData 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.
By Moritz Kirschte · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.