The Big Picture of Generative AI — How VAEs, GANs, Flows & Diffusion Actually Differ (EP40)

3 Просмотры

  • Спасибо! Поделитесь с друзьями!

    Вам не понравилось видео. Спасибо за то что поделились своим мнением!

aii
Издатель
Every model so far has looked at data and answered a question about it.
Module 8 starts a new question: can a model look at data and then make MORE of it?
This is the map of every major generative model family before we build each one.

✅ Discriminative vs generative — P(y|x) vs P(x)
✅ Latent space — the compressed world every generative model samples from
✅ VAE — encoder, bottleneck, decoder, probabilistic latent
✅ GAN — generator vs discriminator, the adversarial game
✅ Normalizing Flow — invertible transforms, exact likelihood
✅ Diffusion — noise forward, denoise backward, today’s quality leader
✅ One-shot vs iterative sampling — why diffusion is slow but SOTA
✅ Trade-off table — quality, diversity, speed, stability, likelihood
✅ Timeline — 2013 VAE to 2025 video diffusion
✅ Which family to reach for, and when

⏱ CHAPTERS:
0:00 Module 8 begins — from describing data to creating it
1:05 Discriminative vs generative — two different questions
2:15 Latent space — the compressed world models sample from
3:25 Meet the four families — VAE, GAN, Flow, Diffusion
4:45 How each one generates — one-shot vs iterative
5:55 Trade-off table — quality, speed, stability, likelihood
7:05 The generative timeline — 2013 to 2025
8:20 Recap + challenge
9:00 Outro → EP41 Autoencoders & VAEs

Категория
AI Картинки
Выполните вход или зарегистрируйтесь для отправки комментария.
Комментариев нет.