Forgetting to Witness: Efficient Federated Unlearning with Visible Evaluation
arXiv·medium signal
Wang, Zhu et al. address the growing need for federated learning models to selectively forget specific data in compliance with privacy regulations. The paper introduces both an efficient unlearning mechanism and a visible evaluation framework — solving the dual challenge of actually removing data influence and proving that removal occurred. Relevant as GDPR right-to-erasure enforcement extends to federated ML deployments.