Amnesiac Machine Learning

The Right to be Forgotten is part of the recently enacted General Data Protection Regulation (GDPR) law that affects any data holder has on European Union residents. It gives EU residents ability request deletion their personal data, including training records used train machine learning models. Unfortunately, Deep Neural Network models are vulnerable information leaking attacks such as model inversion which extract class from a trained and membership inference determine presence an example in model's data. If malicious party can mount attack learn private was meant removed, then it implies owner not properly protected user's rights may compliant with GDPR law. In this paper, we present two efficient methods address question how or delete way they while maintaining efficacy. We start by presenting real-world threat shows simply removing insufficient protect users. follow up removal methods, namely Unlearning Amnesiac Unlearning, enable owners themselves against being regulations. provide extensive empirical analysis show these indeed efficient, safe apply, effectively remove learned about sensitive