Adversarial Sample Detection with Gaussian Mixture Conditional Generative Adversarial Networks

Algorithm 1

We train individual semisupervised generative adversarial networks to study normal data distributions in the pretrained target network’s hidden layers.

	Input: train sample into the pretrained target network.
	for each layer do.
	Train individual semisupervised generative adversarial networks in layer l.
	A semisupervised is obtained.
	end for
	return semisupervised