Computation and Communication Efficient Federated Unlearning via On-server Gradient Conflict Mitigation and Expression

Researchers propose FOUL (Federated On-server Unlearning), a new framework for efficiently removing specific participants' data from federated learning models without accessing client data. The approach reduces computational and communication costs while maintaining privacy compliance through a two-stage process that performs unlearning operations on the server side.

• →FOUL framework enables data removal from federated learning models without requiring access to client data, preserving privacy.
• →The two-stage approach includes a learning-to-unlearn preparation phase and on-server knowledge aggregation for efficient unlearning.
• →New evaluation metric 'time-to-forget' measures how quickly models achieve optimal unlearning performance.
• →Extensive testing shows FOUL outperforms traditional retraining methods with significantly reduced time and resource costs.
• →The framework addresses regulatory compliance requirements for data privacy in distributed AI systems.