GAN (Generative Adversarial Network)

A GAN (Generative Adversarial Network) is a machine learning architecture made of two neural networks trained against each other. The generator creates fake samples from random noise, and the discriminator tries to tell real samples from fake ones. As the discriminator improves, the generator is forced to produce more convincing output.

Why it matters

GANs, introduced in 2014, were the leading approach for realistic image generation for several years and are used for face synthesis, image-to-image translation, super-resolution and data augmentation. They are notoriously hard to train: the two networks can fall out of balance, and the generator can collapse to producing a narrow range of outputs, called mode collapse. Diffusion models have since replaced them in many image-generation uses.

Example

Loop:
1. Generator turns noise into a fake image
2. Discriminator sees real and fake images, outputs real/fake scores
3. Discriminator is updated to detect fakes better
4. Generator is updated to fool the discriminator

Common mistakes

• Assuming a falling generator loss means better images. Judge by samples and metrics.
• Ignoring mode collapse and unstable training.
• Using realistic fakes without disclosure where it could mislead people.

Related

• Diffusion model
• Generative AI
• Neural network
• Synthetic data

This page is general information written by our editorial team and updated when documentation changes. Servers and providers differ, so test changes safely first. See our editorial policy and disclaimer.