Introduced on day 46
Generative adversarial network (GAN)
A generator network trained against a discriminator that tries to separate real from generated samples; sampling takes one forward pass and there is no likelihood.
In detail
The two networks play a minimax game: the discriminator learns to classify real versus generated, the generator learns to produce samples that the discriminator misclassifies. GANs sample fast but train delicately and can cover only part of the data (dropped modes), and without a likelihood a held-out score cannot compare them with other families.