Provable Robustness against Backdoor Attacks via the Primal-Dual Perspective on Differential Privacy
Aman Saxena, Jan Schuchardt, Yan Scholten, Stephan Günnemann {a.saxena, y.scholten, s.guennemann}@tum.de , [email protected] · 2026
Read Provable Robustness against Backdoor Attacks via the Primal-Dual Perspective on Differential Privacy by Aman Saxena, Jan Schuchardt, Yan Scholten, Stephan Günnemann {a.saxena, y.scholten, s.guennemann}@tum.de , [email protected] (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-21780.txt.
Description
Aman Saxena and co-authors connect randomized smoothing to privacy's dual view, giving a clean route to certify toughness against backdoor attacks. Backdoors perturb training and test data jointly, which forces training-time and test-time mechanisms into a single certificate. Privacy profiles supply a number-based recipe for composing varied tools, yielding tight end-to-end proofs that reuse prior private analyses. The method is used for DP-SGD and Deep Partition Aggregation with smoothing at inference, and tests on MNIST and CIFAR-10 confirm its strength.
Catalog record
| CREATOR | Aman Saxena, Jan Schuchardt, Yan Scholten, Stephan Günnemann {a.saxena, y.scholten, s.guennemann}@tum.de , [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 98496 |
| LICENSE | http://creativecommons.org/licenses/by-sa/4.0/ |
| RIGHTS | CC BY SA |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.21780