Generative Models A to Z — From VAE to Stable Diffusion

Provider

KAIST SW교육센터 IT Academy

Dates

2025-07-07–2025-07-11

Duration

20 hours

Format

Hybrid

What I studied

• vae
• gan
• pix2pix
• diffusion
• stable-diffusion
• pytorch

The sequence I used

I started with learning a data distribution, moved through autoregressive factorization and maximum likelihood, and then connected VAE, GAN, Pix2Pix, and diffusion. The source page exposed only one explicit day heading, so these pages follow my study sequence rather than claiming an official day-by-day timetable.

Original papers I revisited

I checked Auto-Encoding Variational Bayes, Generative Adversarial Nets, Pix2Pix, DDPM, Latent Diffusion, and DiT. The diagrams are my own redraws rather than copied paper figures.

Some Colab outputs were not saved. For those topics I stopped at the equations and code path, and I did not invent images that looked like historical runs.

Five-day study notes

1. Day 1 — Generative Models and Deep-Learning Foundations

   I began with what it means to learn a data distribution, then reviewed tensors and neural-network foundations.

2. Day 2 — Autoregressive Models and MLE

   I factorized joint probabilities into conditionals and connected maximum likelihood with negative log-likelihood.

3. Day 3 — Variational Autoencoders

   I studied why the encoder predicts a mean and variance, then connected reparameterization to reconstruction and KL losses.

4. Day 4 — GANs and Pix2Pix

   I separated the generator and discriminator objectives, then followed how a conditional GAN becomes an image-to-image translator.

5. Day 5 — Diffusion, LDM, and Applications

   I connected forward noise to reverse denoising, then compared where DDPM, latent diffusion, and Stable Diffusion perform their computation.