Cycle-Consistent Adversarial GAN: The Integration of Adversarial Attack and Defense

Table 1

Target model architectures for the MNIST. A is the LENET5. B is the LENET5 with dropout.

A	B
Conv1 (32, 3, 3) + Relu	Conv1 (32, 3, 3) + Relu
Max pooling	Conv1 (32, 3, 3) + Relu
Conv1 (64, 3, 3) + Relu	Max pooling
Max pooling	Dropout (0.5)
FC (200)+Relu	FC (128) + Relu
FC (10)	Dropout (0.5)
	FC (10)