ImpMIA: Leveraging Implicit Bias for Membership Inference Attack

Yuval Golbari, Navve Wasserman, Gal Vardi +1

Determining which data samples were used to train a model, known as Membership Inference Attack (MIA), is a well-studied and important problem with implications on data privacy. So…