generative adversarial network

Terms from Artificial Intelligence: humans at the heart of algorithms

Page numbers are for draft copy at present; they will be replaced with correct numbers when final book is formatted. Chapter numbers are correct and will not change now.

In a generative adversarial network (GAN) involves two deep neural networks: one attempts to generate a realistic image (or other media), and the other acts as a critic deterining if the image looks real. The effect is rather like the methd used in game playing by which networks play against each other and learn to be better and better. As the generator network imporves the critic has to learn to make finer distinctions, which then forces the generator to learn to be yet better.

Defined on page 279

Used on Chap. 8: page 158; Chap. 12: page 279; Chap. 21: page 523

Also known as GAN