Core Concepts
Federated unlearning presents unique challenges in the distributed learning context, requiring tailored mechanisms for effective data removal while maintaining model performance.
Abstract
This paper explores the challenges and trends in federated unlearning, focusing on the complexities of removing data in a distributed learning environment. It categorizes existing methods, compares assumptions, and discusses implications for future research. Key highlights include the need for tailored unlearning mechanisms in federated settings, the difficulties in applying centralized unlearning techniques, and the importance of considering data distribution and security aspects in unlearning processes.
Stats
Federated learning (FL) introduced in 2017
Many techniques developed for unlearning in centralized settings are not trivially applicable in FL
Recent work focuses on developing unlearning mechanisms tailored to FL
Quotes
"The unique complexities of unlearning in the federated setting demonstrate the need for tailored unlearning mechanisms for FL."
"Unlearning using historical information could increase the correctness of the unlearned model, as the unlearned model is likely to converge."