Federated Unlearning

The retrained oracle is defined as a model trained from scratch on all retained datasets after removing the target client. Prior federated unlearning methods include partition-based, model-manipulation, and optimization-based approaches. The deletion request may arise from expired consent, data corrections, or privacy regulations such as GDPR and CCPA. Federated unlearning removes a specific client or dataset contribution from an already trained global federated model.