Utility Optimization of Federated Learning with Differential Privacy

Algorithm 3

Client update.

	Input:, learning rate , batch size , clipping threshold , ,
	Output: Private local weight
(1)	Local round  = 0
(2)	whiledo
(3)	Forward pass()
(4)	//Compute gradient
(5)	//Add noise
(6)	//Clip gradient
(7)	//Apply gradient
(8)	j++
(9)	end
(10)	return