Which 2014 neural-network model introduced by Ian Goodfellow generates images by pitting a generator against a discriminator?

Ian Goodfellow’s 2014 neural-network model that pits a generator against a discriminator is a generative adversarial network.

A generative adversarial network, or GAN, contains two competing models. The generator creates synthetic examples, while the discriminator tries to distinguish generated examples from real training examples. During training, the generator improves by learning to fool the discriminator, and the discriminator improves by detecting artificial outputs.

The original GAN paper was titled “Generative Adversarial Nets” and was co-authored by Goodfellow and colleagues. The approach became influential for image synthesis, image-to-image translation, super-resolution, and other generative tasks.

GANs are not the same as diffusion models, which became prominent later and generate samples through a gradual denoising process. GAN training can also be unstable, with problems such as mode collapse, in which the generator produces limited varieties of outputs rather than representing the full data distribution.