After reading the optional notebook of C2W2 on score-based generative modeling, I’m a little lost as there are a lot of mathematical demonstrations. I wondered if anyone had textbooks/learning references to share to make it a little easier to understand. Thanks!
Link to the notebook : Google Colab
Sorry, when I took the GANs courses, I only looked at the GANs material and didn’t look at any of the optional topics like Variational AutoEncoders and Score-Based Generative models. I just took a look at the notebook and the math is pretty deep there. At the very least, you’d need a solid understanding of differential equations, which is a course that a college math or physics major typically would take in their sophomore or junior year. How much math background do you have?
Note that they give lots of links to other papers and also to Wikipedia pages like this one about the Euler-Maruyama Method and this one about Stochastic Differential Equations. You can start by having a look at those and seeing if any of it makes sense to you.
The other approach would be to just assume that understanding the math is optional and try looking at the sample code and running that to see what kind of outputs it generates and the types of operations that it is doing to implement the math.
The high level point here is that this material is completely optional in that it does not have to do with GANs: it’s just introducing us to a completely different technique that can be used for generating images. You could also do some googling and see if there are any courses that cover Score-Based Generative Models if this topic catches your interest.