Generative Adversarial Network (GAN)
Definition
A Generative Adversarial Network (GAN) is a class of machine learning frameworks where two neural networks, a "Generator" and a "Discriminator," contest with each other in a zero-sum game. The Generator tries to create realistic data (like images), and the Discriminator tries to tell the difference between the real data and the fake data created by the Generator.
Why It Matters
GANs are a powerful type of generative model capable of producing stunningly realistic, and entirely artificial, images, art, and other data. They have been a major driver of progress in creative AI.
Contextual Example
To generate realistic human faces, a GAN's Generator would create a face image, and the Discriminator would compare it to a dataset of real photos. Through training, the Generator gets better at fooling the Discriminator, and the Discriminator gets better at spotting fakes, resulting in the Generator producing highly realistic faces.
Common Misunderstandings
- The two networks are trained simultaneously in an adversarial process.
- GANs are notoriously difficult to train.