Conditional GANs against DNS based unsupervised detection of malicious domains

While Unsupervised Learning methods have been employed in detecting malicious domains using DNS data, it is unknown if these techniques are robust in an adversarial setting. We present a novel method using a conditional GAN, where DNS data could be generated to evade detection, while maintaining adversarial functionality. This could also be employed defensively, by using the discriminator as a detector.