IntraShuffler: Privacy-Preserving Heterogeneous Differential Privacy for Federated Learning

arXiv·medium signalXBlueskyLinkedInCopy link

Proposes IntraShuffler, a framework that allows federated learning clients to select individual privacy budgets while maintaining aggregate privacy guarantees through an intra-client shuffling mechanism. Addresses a practical gap: real-world FL deployments need heterogeneous privacy levels across institutions with different data sensitivity policies. Relevant for teams building federated systems across organizations with varying compliance requirements.