Generative Adversarial Network (GAN)

A class of machine learning frameworks designed to generate new data with the same statistics as the training set, consisting of a generator and a discriminator.

How it works

A GAN consists of two neural networks trained in opposition. The generator takes random noise as input and produces synthetic outputs (e.g., images). The discriminator takes either real or synthetic outputs and tries to classify which is which. The generator is trained to fool the discriminator; the discriminator is trained to not be fooled. This adversarial dynamic drives both networks to improve until the generator produces outputs indistinguishable from real data.

Why it matters

GANs produced the first convincing AI-generated faces, artwork, and synthetic training data, sparking a revolution in generative AI in the late 2010s. They enabled deepfakes, style transfer, and super-resolution. While largely supplanted by diffusion models for image generation, GAN architectures remain important in domains where fast inference is critical — since GANs generate in one forward pass, unlike diffusion models that require dozens of denoising steps.