AI Glossary
Generative Adversarial Network (GAN)
An AI architecture that uses two competing neural networks to generate realistic content such as images and videos.
Generative Adversarial Network (GAN)
Overview
One of the most fascinating ideas in artificial intelligence is that models can learn by competing against one another.
Generative Adversarial Networks, commonly called GANs, are built around this concept.
A GAN consists of two Neural Networks. One network generates content, while the other attempts to determine whether that content is real or artificially generated.
The two networks effectively challenge each other.
As the generator becomes better at creating realistic content, the discriminator becomes better at detecting fake content. Over time, this competition can produce remarkably realistic outputs.
A helpful way to think about a GAN is an art forger and an art expert. The forger attempts to create convincing copies, while the expert attempts to identify fakes. Both improve as the competition continues.
GANs became highly influential in image generation and synthetic media. Although many modern image generation systems now rely on Diffusion Models, GANs remain an important milestone in the history of generative AI.
Why It Matters
GANs demonstrated that AI systems could generate realistic content and helped accelerate progress in generative AI.
Real-World Example
GANs have been used to generate realistic human faces, create synthetic training data, and enhance image quality.