What are Diffusion Models?
Essential reference for ML researchers and engineers building or understanding generative models; provides rigorous mathematical foundations and connects diffusion theory to modern architectures like Stable Diffusion.
AI Summary
A comprehensive technical deep-dive into diffusion models, covering forward and reverse diffusion processes, mathematical parameterizations, conditioning techniques (classifier-guided and classifier-free), and optimization methods. This is Lilian Weng's seminal explainer that bridges foundational theory (stochastic gradient Langevin dynamics, score networks) with practical implementations and recent advances like latent diffusion and consistency models.
Excerpt
[Updated on 2021-09-19: Highly recommend this blog post on score-based generative modeling by Yang Song (author of several key papers in the references)]. [Updated on 2022-08-27: Added classifier-free guidance, GLIDE, unCLIP and Imagen. [Updated on 2022-08-31: Added latent diffusion model. [Updated on 2024-04-13: Added progressive distillation, consistency models, and the Model Architecture section.
