RunwayML Week 4 Class Notes
Taught by Derrick Schultz & Lia Coleman
🔗 Links
📣 Class Announcements
We've been talking about having a end-of-class presentations, maybe 2 weeks out from the end of class, possibly in a livestream. You can go live or send us prerecorded video.
Low-key and totally optional. Just meant to be an end-of-class celebration! We'll discuss this more!
How GANs Train
-
GAN = Generative Adversarial Network
-
Adversarial = Two networks that fight against each other. They're called the Generator, and the Discriminator.

-
Analogy:
- A forger (the generator) and a bank (the discriminator).
- At first, both start off really dumb. The forger will draw up really bad fakes. The bank will reject it, and then the forger will learn how to produce a slightly better fake. Through more and more training steps, both the forger and the bank get better and better. The forger makes better fakes, and the bank will get better at telling fakes from real dollars.




🧐 Why Use Runway (over other Training Options)
✅ Pros