What I noticed a few years ago when I decided to bring myself up to date again on machine learning after a ten years break (and what ten years that was for the domain!) I read a ton and watched several online classes.

I realized that the pace was very different from the classes I had been during my student years: classes were boring, so I had time to think about stuff.

The rhythm that I found worked really well for such information-dense subjects was 1:1 breaks. One hour of classes. One hour to think about it (usually I would go for a walk)

Internet taught us to drink from a firehose, but our brain needs some time to process the information. It can't accumulate information and digest it at the same time.

You just learned about drop-out layers or the drawbacks of softmax? Don't feel bad about switching off the computer and think about it. No one is judging that you are "doing nothing".

I remember that on my last corporate job, I used to go for a break/walk when stuck on a nasty bug and would often come back with the solution. My colleagues frowned a bit upon that but luckily my boss, a former researcher, totally approved of the method.