This study delves into the concept of ‘the right to be forgotten’ in line with data protection regulations like the GDPR. It focuses on Federated Online Learning to Rank (FOLTR) systems where a ranker is learned by aggregating local updates to a global ranking model. The paper presents an efficient unlearning method to erase a client’s contribution to the learning model without compromising its effectiveness or having to retrain it from scratch. The effectiveness of this unlearning strategy is tested on four datasets, under different parameter settings.
Publication date: 24 Jan 2024
Project Page: https://arxiv.org/abs/2401.13410v1
Paper: https://arxiv.org/pdf/2401.13410
