generativedeep-learning
Diffusion and score-based models
DDPM, score matching, flow matching — SOTA generative models behind Stable Diffusion, Sora, etc.
Уровни глубины
L0Intro~2ч
Knows diffusion = "add noise step-by-step, then learn to reverse".
L1Basics~15ч
Implements DDPM forward / reverse processes on toy data; noise schedules.
L2Working~25ч
Trains a conditional diffusion model with classifier-free guidance; uses DDIM sampling.
L3Advanced~35ч
Score-SDE formulation; consistency models, flow matching, latent diffusion.
L4Research~80ч
Continuous-time theory, fast solvers, diffusion for non-image modalities.
Ресурсы
L1 — Basics