To generate malware samples to attack machine learning models, attackers are now using GANs to achieve their goals. Using the same techniques we discussed previously (a generator and a discriminator), cyber criminals perform attacks against next-generation anti-malware systems, even without knowing the machine learning technique used (black box attacks). One of these techniques is MalGAN, which was presented in a research project called, Generating Adversarial Malware Examples for Black Box Attacks Based on GAN, conducted by Weiwei Hu and Ying Tan from the Key Laboratory of Machine Perception (MOE) and the Department of Machine Intelligence. The architecture of MalGAN is as follows:
The generator creates adversarial malware samples by taking malware (feature vector m) and a noise vector, z, as input. The substitute detector is a multilayer, feed-forward neural network...
