AI Glossary Term

What does “GAN (Generative Adversarial Network)” mean?

Definition

A generative setup where a generator and discriminator train against each other.

Browse the full AI GlossaryExplore all free AI guides

How GAN (Generative Adversarial Network) works

Diagram for GAN (Generative Adversarial Network), a repeating loop: Generator makes a fake; then Discriminator compares fake vs real; then Discriminator says which is which; then Both improve, repeat; then back to the start.Generator makes a fakeDiscriminator compares fake vsrealDiscriminator says which is whichBoth improve, repeatrepeat

Example

In an early GAN face generator, one network invents faces while another tries to tell them from photos. As the detector gets better, the generator must produce more convincing faces to fool it.

Why it matters

GANs produced some of the first very realistic synthetic images and helped start public debate about deepfakes. Diffusion models have since overtaken them in many uses.

Common questions

Who invented GANs?

Ian Goodfellow and colleagues introduced them in 2014.

Why are GANs hard to train?

The two networks must stay balanced. If one overpowers the other, training can stall or the generator may repeat a narrow set of outputs, called mode collapse.

Are GANs still used?

Yes, in areas such as image enhancement and synthetic data, though diffusion models are now more common for general image generation.

Related terms

Latent SpaceA compressed representational space where similar concepts are positioned near each other as vectors.Diffusion ModelA generative architecture that learns to reverse noise to synthesize images, audio, or other content.Sliding Window AttentionAn attention pattern where each token attends only to a fixed-size window of nearby tokens to reduce compute.Self-PlayA training setup where a model improves by generating data through interactions or competitions with copies of itself.Generative AIAI systems that produce new content such as text, images, audio, video, or code.GradientA vector showing how much each parameter should change to reduce loss.

Learn more in our free guides

Neural NetworksGenerative AIHard Parameter Sharing in Multi-Task NetworksBidirectional Recurrent Networks

Test yourself

Neural Networks QuizGenerative AI QuizHard Parameter Sharing in Multi-Task Networks Quiz

See also

Function CallingGeneralizationFoundation ModelFine-TuningFew-Shot LearningGradient DescentFeature ExtractionGround Truth