Emerging Challenges and Future Directions in Federated Unlearning

Federated Unlearning (FU) is becoming increasingly important in Federated Learning (FL) for safeguarding privacy by enabling the removal of specific client data from trained models. This review delves into the key challenges and emerging directions in FU, particularly around incentive mechanisms, environmental sustainability, and its use in large foundation models. We explore how personalized incentives can help retain users and examine strategies to reduce energy consumption in unlearning processes, promoting greener AI. Additionally, we address the complexities of implementing FU in large-scale models, where partial retraining may cause unpredictable impacts on model performance. Tackling these issues is crucial for advancing FU solutions that are scalable, efficient, and sustainable, especially as FL expands across diverse applications.