“The integration of machine learning (ML) in numerous critical applications introduces a range of privacy concerns for individuals who provide their datasets for model training. One such privacy risk is Membership Inference (MI), in which an attacker seeks to determine whether a particular data sample was included in the training dataset. … MI attacks capitalize on access to the model’s predicted confidence scores to successfully perform membership inference, and employ data poisoning to further enhance their effectiveness. … We show that existing label-only MI attacks are ineffective at inferring membership.”

Find the paper and full list of authors at ArXiv.