On the Difficulty of Membership Inference Attacks

Shahbaz Rezaei, Xin Liu

Recent studies propose membership inference (MI) attacks on deep models, where the goal is to infer if a sample has been used in the training process. Despite their apparent succes…