Evaluations of Machine Learning Privacy Defenses are Misleading
This work exposes misleading practices in privacy evaluations, which is critical for researchers and practitioners relying on empirical defenses for machine learning.
The paper identifies severe pitfalls in existing evaluations of machine learning privacy defenses, showing they underestimate privacy leakage by an order of magnitude in 5 case studies, and finds that none of these defenses outperform a properly tuned DP-SGD baseline.
Empirical defenses for machine learning privacy forgo the provable guarantees of differential privacy in the hope of achieving higher utility while resisting realistic adversaries. We identify severe pitfalls in existing empirical privacy evaluations (based on membership inference attacks) that result in misleading conclusions. In particular, we show that prior evaluations fail to characterize the privacy leakage of the most vulnerable samples, use weak attacks, and avoid comparisons with practical differential privacy baselines. In 5 case studies of empirical privacy defenses, we find that prior evaluations underestimate privacy leakage by an order of magnitude. Under our stronger evaluation, none of the empirical defenses we study are competitive with a properly tuned, high-utility DP-SGD baseline (with vacuous provable guarantees).