Skip to content
← All tracks

Generative Models

PyTorch

From autoencoders through VAEs to modern diffusion. The math behind generating images, audio, and text.

0 / 11 solved
Continue →
  1. 1. Not solved yet. Autoencoder
  2. 2. Not solved yet. Simple GAN Generator
  3. 3. Not solved yet. VAE Reparameterization Trick
  4. 4. Not solved yet. VAE ELBO Loss
  5. 5. Not solved yet. DDPM Noise Schedule
  6. 6. Not solved yet. DDPM Forward Noising
  7. 7. Not solved yet. DDPM Denoising Step
  8. 8. Not solved yet. Classifier-Free Guidance
  9. 9. Not solved yet. Train VAE End-to-End
  10. 10. Not solved yet. DDPM Training Step End-to-End
  11. 11. Not solved yet. DDPM Sampler Loop

Check yourself

4 questions · one attempt each

These do not count toward finishing the track. They are here to catch the things that are easy to read past.

0 / 4

VAEs tend to produce blurry samples and GANs sharp ones. What explains the difference? 

# VAE: maximise a likelihood bound under a pixel-wise reconstruction loss
# GAN: minimise a discriminator's ability to tell real from fake
Question 1 of 4