GAN

TL;DR

Composed of two models trained adversarially: one generates synthetic data and the other discriminates between real and fake.
The two parts improve together: the generator produces increasingly realistic outputs while the discriminator becomes better at spotting fakes.
Commonly used to produce realistic images or text, augment datasets, generate test data, and support data privacy or anomaly detection.

Definition

A GAN, or Generative Adversarial Network, is a type of deep learning algorithm that consists of two parts: a generator and a discriminator. The generator creates fake data that is meant to be indistinguishable from real data, while the discriminator attempts to differentiate between the real and fake data.

Explanation

The generator and discriminator are trained together in an adversarial process. The generator’s goal is to produce synthetic examples that the discriminator classifies as real; the discriminator’s goal is to correctly identify whether inputs are real or generated. Through this competitive training, the generator learns to create more realistic data and the discriminator becomes better at identifying fakes. GANs are therefore a powerful tool for producing synthetic data that closely resembles real data.

Examples

Computer vision

The generator creates fake images of objects, such as cars or animals, while the discriminator is trained on a dataset of real images. The generator and discriminator work together to improve their performance, with the generator learning to create more realistic images and the discriminator becoming better at identifying fake images.

Natural language processing

The generator creates fake text, such as fake news articles or social media posts, while the discriminator is trained on a dataset of real text. The generator and discriminator work together to improve their performance, with the generator learning to generate more realistic text and the discriminator becoming better at identifying fake text.

Use cases

Creating realistic images or text for use in training other machine learning algorithms
Generating fake data for testing and evaluation
Data augmentation
Data privacy
Anomaly detection

Generator (model)
Discriminator (model)
Deep learning
Synthetic data

GAN