← Deep Learning learning pathDeep Learning · Module 9 of 10
Generative models
Create new images, text and audio with generative AI.
📖 Key ideas
- Generative models learn to produce new data.
- GANs: a generator competes with a discriminator.
- VAEs learn a compressed latent space.
- Diffusion models generate images by removing noise step by step.
- LLMs generate text one token at a time.
💡 Real-life example: A GAN is a forger and a detective improving each other until fakes look real.
GANGenerative Adversarial Network
Diffusion modelGenerates by denoising
Latent spaceCompressed representation
TokenUnit of text for LLMs