[NeurIPS 2024] Official implementation of the paper “Ferrari: Federated Feature Unlearning via Optimizing Feature Sensitivity"

An Empirical Study of Federated Unlearning: Efficiency and Effectiveness (Accepted Conference Track Papers at ACML 2023)

ConDa is an efficient federated unlearning framework that removes a client's data from a global model without retraining or additional computational overhead. It outperforms existing methods by at least 100× in non-IID federated learning settings.