L2 regularization lowers membership inference attack advantage on some datasets, but the effect is absent on MNIST, and the comparison with differential privacy is under-specified.
In: Proceedings of the 2016 ACM SIGSAC conference on computer and communications security
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Effectiveness of L2 Regularization in Privacy-Preserving Machine Learning
L2 regularization lowers membership inference attack advantage on some datasets, but the effect is absent on MNIST, and the comparison with differential privacy is under-specified.