Adversarial Sample Detection with Gaussian Mixture Conditional Generative Adversarial Networks
Algorithm 2
In the supervised and partially supervised scenario, the final detection classifier is the support vector machine classifier. We used 10% of the test set as training data and 90% as evaluation data.
Input: the normal data and the adversarial samples into the pretrained network.
for each layer do
Input: the hidden layer l into semisupervised (Algorithm 1)
is obtained.
errori is obtained.
end for
The normal data x latent feature is and the logit vector of the target network.
The adversarial samples latent feature is and the logit vector of the target network.
We use the normal data x latent feature and adversarial samples latent feature input the support vector machine classifier.
