Aegis Protects Medical Federated Learning via Generative Gradient Masking

Aegis provides a client-side defense against model inversion attacks in federated learning without the accuracy trade-offs typical of differential privacy. It prevents malicious servers from reconstructing private patient images from shared model updates while maintaining clinical diagnostic reliability.