Federated Unlearning: Techniques, Trends, and Future Directions
Hussien AbdelRaouf, Mohamed I. Ibrahem · 2025
Federated Unlearning (FU) has emerged as a crucial framework to address privacy and regulatory compliance in Federated Learning (FL) systems. FU enables the removal of specific data contributions from trained models while preserving overall performance and utility. This survey explores FU methodologies, categorizing them into class-level, sample-level, and client-level unlearning, and examines recent techniques, including retraining-based, gradient-based, and parameter editing approaches. Emerging trends such as blockchain integration for enhanced transparency and incentive mechanisms for client participation are discussed. The paper identifies key challenges, including the complexity of applying FU to large-scale models, maintaining client engagement, and efficiently addressing multiple unlearning requests. It also proposes future research directions to develop robust, scalable, and privacy-preserving FL systems.