DataStealing

Federated Learning (FL) iscommonly used tocollaborativelytrain models with privacypreservation. Specifically,AdaSCP evaluates the importance of parameters with the gradients in dominant timesteps of the diffusion model.