Generative Models
PreviewVAEs, GANs, flows and diffusion — derived, not hand-waved.
Start the first topicHow models learn to produce data. Every topic is a derivation, an intuition and short checks: first why a construction is needed, then where it comes from, then whether you can use it. The course is built from open seminar notes; the full notes stay free to read.
Curriculum
1 of 15 topics available01Foundations
- What generative modeling isPlanned
- Multivariate GaussiansPlanned
- Entropy, cross-entropy and KLPlanned
- AutoencodersPlanned
02Variational autoencoders
- Latent variables and the ELBOPlanned
- The reparameterization trick PracticeAvailable
- β-VAE and VQ-VAEExtensionPlanned
03Generative adversarial networks
- GANs as a two-player gamePlanned
- Wasserstein GANs and Lipschitz constraintsPlanned
04Normalizing flows
- The change of variablesPlanned
05Diffusion models
- DDPM: the forward processPlanned
- DDPM: reverse process and objectivePlanned
- Score matchingPlanned
- Latent diffusionPlanned
06Evaluation
- Fréchet Inception DistancePlanned