Fast and Accurate Deep Leakage from Gradients Based on Wasserstein Distance

Algorithm 1

WDLG Algorithm.

	Input: : real gradient generated by real training sample . Epochs: maximum number of iterations. : learning rate.
(1)	//Get the subscript of the last layer of negative bias
(2)	One_Hot //Convert to One_Hot code
(3)	//Initialize virtual data with the same dimensions as
(4)	for to do
(5)	//Calculating virtual gradients
(6)	//WDCA is Algorithm 2
(7)
(8)	end for
	Output: