Forgetting by Design

Testing the Effectiveness of Machine Unlearning in Right to Be Forgotten Data Deletion

Abstract

The Right to Be Forgotten (RTBF) is a legal requirement that, when implemented, allows individuals to request their information be deleted from digital media. However, with the use of machine learning models in today’s modern age, full data deletion can be technically challenging. In this research paper, the author will evaluate the effectiveness of machine unlearning as a complete data-deletion method for complying with RTBF. The research method uses a pre-trained neural network, tested with varying sizes of a forget set and Membership Inference Attacks (MIA), to determine whether the model retains information from deleted data. The results highlight the  limitations of machine unlearning and the implications it can have on RTBF.