Password Guessing Based on GAN with Gumbel-Softmax

Algorithm 1

G-Pass with gradient penalty.

(i)	Input: The gradient penalty coefficient , the maximum of training iterations , the number of discriminator iterations per generator iteration , the batch size , the learning rate , Adam hyperparameters and , initial discriminator parameters , initial generator parameters , the highest temperature
(1)	for to do
	/^∗ training discriminator ^∗/
(2)	for to do
(3)	for to do
(4)	Sample real data , , random number
	, is the temperature of current iteration;
(5)	;
(6)	;
(7)	;
(8)	end
(9)	;
(10)	end
	/^∗training generator ^∗/
(11)	Sample a batch of input of generator ;
(12)
	/^∗ update the temperature ^∗/
(13)
(14)	end