Approach

Technique

Main Idea

Dwork et al. [52]

Binomial mechanism

Propose a Binomial mechanism.

Agarwal et al. [53]

FL + Binomial mechanism

Apply the binomial mechanism to federated learning and propose a stochastic k-level quantization method and a randomized rotation method to optimize the communication efficiency.

Canonne et al. [54]

Discrete Gaussian mechanism

Propose a discrete Gaussian mechanism.

Kairouz et al. [55]

FL + Discrete Gaussian mechanism

Apply the discrete Gaussian mechanism to federated learning, and results show that the model can provide $1/2{\epsilon }^2$ central differential privacy.

Agarwal et al. [56]

FL + Skellam mechanism

Proposes a Skellam mechanism based on the addition of the difference of two independent Poisson random variables; Apply it to federated learning and achieve $\left(\alpha \mathrm{,}\epsilon \left(\alpha \mathrm{<mo\; stretchy="false">\; )\; </mo>}\right)\right)$ Rényi differential privacy.

Bao et al. [57]

FL + Skellam mixture mechanism

Proposes a Skellam mixture mechanism based on a mixture of two shifted symmetric Skellam distributions; Apply it to federated learning and achieve (ε − δ) differential privacy condition.

Chen et al. [58]

FL + Poisson binomial mechanism

Proposes a Poisson binomial mechanism; Apply it to federated learning and achieve the (ε − δ)-approximate differential privacy.