Weighted data reconstruction attack in Federated Learning
November 13, 2026
An interpolation-based approximation method allows attackers to reconstruct training data in FedAvg scenarios by generating intermediate model updates. The attack uses a layer-wise weighted loss function to improve reconstruction quality.
HOW THIS AFFECTS YOU
●
researcherYou can use this weighted loss approach to test the robustness of your federated learning implementations.
●
policyYou should be aware that standard Federated Averaging may not be as private as assumed against reconstruction attacks.