Beating The Fakers As Generative Adversarial Networks Rise

GAN, a technique in unsupervised ML that pits two neural networks against each other in a zero-sum game framework, has a variety of applications including autonomous driving, text generation, video generation and digital image repair. Machine-learning ethicists and practitioners have begun to raise concerns over their application however, as computer-generated images, video and audio become more realistic, while still stressing that the tools could open up vital new areas of research for the ML community. Already, GANs are able to generate pictures of faces that look convincingly real to the untrained eye, Tamir explained. Earlier this month, OpenAI made a decision to not open-source their GAN based GPT-2 text-generation model, citing ethical concerns over the possible misuse of such technology in generating fake news articles. Already, GANs are able to generate pictures of faces that look convincingly real to the untrained eye.