Find out what the critic rewards

A journey through generative adversarial networks with Botie.

Generative Models · A journey with Botie

First visit? Learn the workshop controls

Botie's notebook

Could Another Model Tell the Generator What Looks Wrong?

Part 1 of 3

Writing a formula for whether an image looks plausible is difficult. Pixel-by-pixel agreement with one target is also a poor description of all the different images that could be valid. We would like a learning signal that can adapt to the kinds of mistakes a generator makes.

Sources for this history

Goodfellow and colleagues (2014), Generative Adversarial Networks

← Previous lesson: Variational Autoencoders Read the full lesson and derivations