Approach

Technique

Main Idea

Geyer et al. [28]

FL + CDP

Initiate the study of federated learning with central differential privacy and verify the validity of the model by numerical experiments

Triastcyn and Faltings [29]

Bayesian differential privacy

Develop a relaxation of federated learning with central differential privacy, named Bayesian differential privacy.

Wei et al. [30]

NbAFL

Develop a novel framework (named NbAFL) based on DP in which artificial noises are added

Zhang et al. [32]

Clipping-enabled FedAvg

Develop a novel central differential privacy framework (named clipping-enabled FedAvg) based on clipping technique.

Hu et al. [33]

Fed-SMP

Develop a novel framework, named Fed-SMP, to mitigate the inaccuracy issue of LDP by using a technique called Sparsified Model Perturbation (SMP) where local models are sparsified first before being perturbed by Gaussian noise.