Definition
Plain language
A way of making AI images by pitting two networks against each other, one creating fakes and one trying to spot them.
As stated in the literature
Generative adversarial network; a generator and discriminator trained in opposition, the generator learning to produce samples the discriminator cannot distinguish from real data.
Also called: GANs, generative adversarial network
Why it matters: It offers a way to generate convincing new images and data by turning training into a contest, though the competing setup can be tricky to keep balanced.
For example, one network invents a face while the other guesses whether it is real, and the inventor keeps improving until the guesser can no longer tell.
Heard on the show
“One slide from a GAN paper went from content overflowing the frame by more than half its height, six defects, down to zero.”Episode 217 — Why an AI Called Fourteen Broken Figures Perfect, And What It Reveals About Test-Time Compute