Lessons
1
The generation problem: modeling p(x)
2
Autoencoders: compression and reconstruction
3
Variational autoencoders: stochastic encoders
4
The ELBO: deriving the VAE objective
5
The reparameterization trick
6
GANs: the minimax game
7
GAN training dynamics: mode collapse
8
Diffusion models: the forward noising process
9
Reverse diffusion and DDPM
10
Score matching: the deeper theory