Note: I realize this response sounds similar to what you have already tried, but it's slightly different in a way that might be useful to you.
Background: senior year of college I took one of the deep learning courses my CS dept was offering, and I was nervous about it early on because I nearly failed the math pre-req test they gave to assess whether we could handle the concepts in the course and get some value out of it.
I ended up doing well in the course and passed with a great grade, even though I felt like I was drinking from a technical firehose the whole time. The technique I ended up using to get myself through was what I call depth-first learning. Basically you start from the beginning of a non-trivial, legitimate starter project or material, and work through it as you normally would. When you hit a point of confusion or a knowledge gap of any size, pause everything and follow the concept graph all the way down until you bottom out or hit something you have seen before and understand, then backtrack and continue from where you first branched. Taking notes in a gdoc or something is also crucial.
I'm not claiming to have reverse-engineered cold fusion or anything by learning this way, but my main takeaway was this: everything I thought was complex was really just a huge pile of simple things that I understand (or could understand with minimal effort). It was all about understanding the essence of what is going in the big picture and how the smaller pieces relate to it. The main difference I think in this approach vs what you have already tried is that you are using a similar approach with a reference manual, and not a legit near-state-of-the-art project with a concrete goal in mind. If you don't try to accomplish something specifically, you're just going to load up a bunch of information in your head, then eject it when you come across a problem you are actively trying to solve.
Background: senior year of college I took one of the deep learning courses my CS dept was offering, and I was nervous about it early on because I nearly failed the math pre-req test they gave to assess whether we could handle the concepts in the course and get some value out of it.
I ended up doing well in the course and passed with a great grade, even though I felt like I was drinking from a technical firehose the whole time. The technique I ended up using to get myself through was what I call depth-first learning. Basically you start from the beginning of a non-trivial, legitimate starter project or material, and work through it as you normally would. When you hit a point of confusion or a knowledge gap of any size, pause everything and follow the concept graph all the way down until you bottom out or hit something you have seen before and understand, then backtrack and continue from where you first branched. Taking notes in a gdoc or something is also crucial.
I'm not claiming to have reverse-engineered cold fusion or anything by learning this way, but my main takeaway was this: everything I thought was complex was really just a huge pile of simple things that I understand (or could understand with minimal effort). It was all about understanding the essence of what is going in the big picture and how the smaller pieces relate to it. The main difference I think in this approach vs what you have already tried is that you are using a similar approach with a reference manual, and not a legit near-state-of-the-art project with a concrete goal in mind. If you don't try to accomplish something specifically, you're just going to load up a bunch of information in your head, then eject it when you come across a problem you are actively trying to solve.