Membership Inference Attack

A privacy attack paradigm that aims to determine whether a specific data sample appeared in the training set of a target model. It has traditionally targeted model weights or output probabilities.