TAMIS: Tailored Membership Inference Attacks on Synthetic Data
This work addresses privacy concerns for users of synthetic data generation methods, but it is incremental as it builds upon an existing attack.
The paper tackles the problem of assessing privacy in differentially-private synthetic data generation by proposing TAMIS, a tailored membership inference attack that reduces computational cost and attacker knowledge compared to the state-of-the-art MAMA-MIA, achieving similar or better performance in experiments on SNAKE challenge replicas.
Membership Inference Attacks (MIA) enable to empirically assess the privacy of a machine learning algorithm. In this paper, we propose TAMIS, a novel MIA against differentially-private synthetic data generation methods that rely on graphical models. This attack builds upon MAMA-MIA, a recently-published state-of-the-art method. It lowers its computational cost and requires less attacker knowledge. Our attack is the product of a two-fold improvement. First, we recover the graphical model having generated a synthetic dataset by using solely that dataset, rather than shadow-modeling over an auxiliary one. This proves less costly and more performant. Second, we introduce a more mathematically-grounded attack score, that provides a natural threshold for binary predictions. In our experiments, TAMIS achieves better or similar performance as MAMA-MIA on replicas of the SNAKE challenge.