What does GAN stand for in artificial intelligence?

GAN stands for generative adversarial network in artificial intelligence.

Ian Goodfellow and collaborators introduced the GAN framework in a 2014 paper. It uses two neural networks with competing jobs: a generator creates synthetic examples, while a discriminator tries to distinguish those examples from real training data.

During training, the generator improves by attempting to fool the discriminator, and the discriminator improves by detecting generated samples. This adversarial process can produce realistic images, audio, video, and other data. The approach became an important milestone in generative AI before the rise of large diffusion models.

The word “adversarial” does not mean that the system is hostile in a human sense. It describes the training contest between the two networks. GANs are also distinct from ordinary classifiers: a classifier predicts categories, whereas a GAN is designed primarily to learn how to generate new examples.