Adversarial Networks

Adversarial network is an implementation of generative adversarial network, which is used to generate adversarial samples in batches for a specified neural network model.

A trained GAN can effectively generate a large number of different adversarial examples. This can be used by attackers to generate attacks that are different from previous attacks. But defenders can also generate labeled negative inputs to enhance the training of their classifiers.

In early 2018, it was proposed by researchers from Carnegie Mellon University and the University of North Carolina at Chapel Hill.