Abstract: Privacy-preserving machine learning requires that models do not reveal any private information about their training data. However, model inversion attacks ...