Diffusion models for imaging and vision Just made this set of slides for an upcoming talk. Feedback is welcome! Slides: Tutorial (March version): (Thx for messaging me the typos. Still working on them.)

Dark mode for this paper for night readers 🌙

Diffusion models are just long VAEs

it's a very intuitive and beginner-friendly explanation, loved it! 👏 I'd suggest adding more explanation about Gaussian and it's importance. it's a topic often overlooked

Great work! Could I watch the lecture on YouTube? Thank you!

Interesting to consider diffusion beyond image gen check out a paper summary and QA here:

Nice formula-writeup. Some grammar issues. Feedback: In writing a cooking recipy or image generator, there's freedom for arbitrary choices. But Langevin dynamics is a model of physics. Sure you can re-run the math it's captured in (simulate), but it's is not an algorithm itself.

There is a mistake in the proof of equation (32) in the paper: Tutorial on Diffusion Models for Imaging and Vision. The last expectation should be E_{q_{\phi}(\vx_{t}|\vx_{0})}.