Any recommended material to learn about the rational behind diffusion models like DDPM and other noise-based generators?

Scientist and engineer interested in Life.

Work on (GOF)AI, Agriculture, Biology, Energy, Robotics, Space---all topics climate-aware.

Kuriyama, Yubari District, Hokkaido, Japan

I came across research that casts noise-based diffusion as a model of biological evolution (from Michael Levin’s group, unsurprisingly). And it is for now a better intuition, still feeling it lacks a more generic sense.

The paper is fine, as well as some posts that explain more of the paper or walk through detailed implementation.

But contents so far seem to eschew explaining more intuition from sampling noise a smart way to “get back” learned content.