GANs (Generative Adversarial Networks)

GANs consist of two neural networks competing against each other to generate increasingly realistic outputs.

• **Architecture:**
• Generator: Creates fake samples
• Discriminator: Tries to distinguish real from fake
• They train together, improving each other

How It Works: 1. Generator creates fake image 2. Discriminator guesses if it's real or fake 3. Both networks update based on results 4. Generator gets better at fooling discriminator

Applications: - Image generation (before diffusion dominated) - Face generation (StyleGAN) - Image-to-image translation - Super resolution

Limitations: - Training instability - Mode collapse - Now largely superseded by diffusion models