Adversarial learning/training pitches two or more algorithms against each other in a competitive situation. Each adversary uses machine learning to attempt to gain advantage over the other. It takes inspiration from zero-sum games, where one player's gains are the other player's losses. The technique has been used in generative adversarial networks for image generation, for example in deep fakes, where one adversary tries to generate new images and the other to distinguish generated images from real ones. Self play is a similar technique in game playing AI.
Also known as: adversarial, adversarial techniques, adversarial training
Used in glossary entries: deepfake, game playing, generative adversarial network, machine learning, self play