Paper 2026/612
Improving ML Attacks on LWE with Data Repetition and Stepwise Regression
, FAIR at Meta, Duke University, Axiom MathAbstract
The Learning with Errors (LWE) problem is a hard math problem in lattice-based cryptography. In the simplest case of binary secrets, it is the subset sum problem, with error. Effective ML attacks on LWE were demonstrated in the case of binary, ternary, and small secrets, succeeding on fairly sparse secrets. The ML attacks recover secrets with up to 3 active bits in the "cruel region" (Nolte et al. 2024) on samples pre-processed with BKZ. We show that using larger training sets and repeated examples enables recovery of denser secrets. Empirically, we observe a power-law relationship between model-based attempts to recover the secrets, dataset size, and repeated examples. We introduce a stepwise regression technique to recover the "cool bits" of the secret.
Metadata
- Available format(s)
-
PDF
- Category
- Attacks and cryptanalysis
- Publication info
- Preprint.
- Keywords
- learning with errorscryptanalysismachine learning
- Contact author(s)
-
eshika @ meta com
klauter @ meta com - History
- 2026-03-30: approved
- 2026-03-27: received
- See all versions
- Short URL
- https://ia.cr/2026/612
- License
-
CC BY
BibTeX
@misc{cryptoeprint:2026/612,
author = {Alberto Alfarano and Eshika Saxena and Emily Wenger and François Charton and Kristin Lauter},
title = {Improving {ML} Attacks on {LWE} with Data Repetition and Stepwise Regression},
howpublished = {Cryptology {ePrint} Archive, Paper 2026/612},
year = {2026},
url = {https://eprint.iacr.org/2026/612}
}
