no code implementations • 4 Mar 2022 • Guoyao Li, Shahbaz Rezaei, Xin Liu
In this paper, we develop a user-level MI attack where the goal is to find if any sample from the target user has been used during training even when no exact training sample is available to the attacker.