Broad intuition · Study map 21 of 85
VAEs, GANs, diffusion, and flow matching
Suggested preparation: Graphs and recommenders: learning from relationships
Picture it
Generative models learn ways to move from simple randomness to structured examples. A VAE learns a compressed latent space; a GAN learns against a critic; diffusion repeatedly denoises; flow matching learns a velocity field that transports samples.
What to understand
- Latent variables and reconstruction
- Adversarial learning, denoising, sampling steps, and flow trajectories
Follow the references
1 · Course
Hugging Face · Introduction to diffusion models ↗Follow the noise-to-image notebook. Change the number of sampling steps and inspect the outputs.
2 · Paper
Lipman et al. · Flow matching for generative modeling ↗Read the introductory path/velocity picture; leave the derivation for later.
You understand it when…
Why does cutting the number of generation steps change both latency and quality?
This is a study map. Follow the reference for the full explanation; use the question to test your mental model. Browse the reference shelf.