I’m curious if you’ve looked at Learning From Data by Abu-Mostafa, Magdon-Ismail, and Lin? (There’s also a lecture series from CalTech based off the book.)
I haven’t read Understanding Machine Learning, but it does seem to be an even more technical, given my skimming of your notes. However, the Mostafa et al book does give a proof of why you can expect the VC dimension to be polynomially bounded for a set of points greater than the break point (if the VC dimension is finite), as well as a full proof of the VC Bound in the appendix.
I haven’t. Would you say that it is good?
UML also includes the proof though – it hasn’t skipped any important proof so far. I just didn’t want to replicate it here, because I didn’t feel like understanding the steps of the proof really helped my intuition in any significant way.
LFD was my first intro to statistical learning theory, and I think it’s pretty clear. It doesn’t cover the No Free Lunch Theorem or Uniform Convergence, though, so your review actually got me wanting to read UML. I think that if you’re already getting the rigor from UML, you probably won’t get too much out of LFD.