A Survey on Federated Unlearning: Challenges, Methods, and Future Directions

The "Right to Be Forgotten" is becoming legally and ethically essential: The paper emphasizes that data privacy regulations now require mechanisms for individuals to request removal of their personal data from AI systems, making federated unlearning a practical necessity rather than just a theoretical concept for digital trust and AI safety.

Federated unlearning addresses a unique privacy challenge in distributed AI: Unlike traditional machine unlearning, federated unlearning must handle data erasure across multiple independent devices or organizations that collectively train a shared model, requiring specialized technical approaches that account for the decentralized nature of federated learning systems.

Current federated unlearning methods face significant technical obstacles: The survey identifies that the distinctive characteristics of federated learning—such as data distribution across multiple parties and limited central control—create specific challenges that standard unlearning techniques cannot adequately address, necessitating tailored algorithm designs.

The field lacks standardized approaches and clear best practices: The paper notes that while various federated unlearning schemes exist, there is no unified workflow or widely accepted methodology, suggesting the field is still developing consensus on how to effectively implement these privacy protections in practice.

Real-world applications remain limited and face practical constraints: The survey indicates that despite theoretical progress, practical applications of federated unlearning are constrained by various limitations, highlighting a gap between what researchers have developed and what organizations can reliably deploy today.