Make It Stick: The Science of Successful Learning
This book (link) is a good refresher on the science of learning. While the principles applied here are rather straightforward, we - or at least I - often tend to forget why they are correct, and how important they are. What makes the book special compared to others in this field, is their emphasis that the best learning techniques often do not feel good and effective (they are hard), while the bad ones feel good (they are easy).
My incomplete takeaways
The main message of Make it stick is this: The learning techniques x and y are good, even though they feel bad. Everything else elaborates on this.
What works and what does not?
The main techniques that work are: Active retrieval, spaced repetition and interleaving. Their not-working, widespread counterparts are re-reading, bulk learning and "one focus" learning (you only learn about one thing at a time).
We think that re-reading works, but really it brings only an illusion of knowledge through familiarity. Bulk learning, also called cramming when done before exams, does the same. And for a very short time after, we can remember the things rather well. But only shortly after, almost nothing is retained. For interleaving, we indeed do a little bit worse than "one focus" shortly after, but after a bit of passed time, we do more than two times better. Interleaving can mean mixing up different subjects, mixing up parameters, including random selections and other variations. They all help because they make us see the bigger picture.
Learning should be hard. Thus, struggling is not a sign of failure, but of progress. You should struggle!
In the book, this is referred to as desirable difficulties. Essentially, when you have to use more effort, the learning will be better. So you should really think hard. Examples include: Thinking about what will be said in the text/ lecture beforehand, testing what you remember about the text/ lecture afterwards (without looking it up), and even just literal thinking about a topic. As in, ask yourself this question right now: "How does the transformer model work?" And if you can not explain that right now, then you probably will neither at test time. Again, you might feel that you know how the transformer model works. You probably read a lot about it and by now feel really familiar with the content. But that does not mean that you can produce it!
Additional concepts
There are a few other concepts that will improve your learning. The first one is elaboration. Everything you know is connected to cues. And only when those cues are sufficiently triggered, can you retrieve that knowledge. A good example is remembering how your childhood feels. Alone, this might be hard. But now think about going back to your hometown, and those memories will come flooding back. The place and the feeling are connected. In the same way, many other cues can help you strengthen and increase your connections to certain knowledge. Good examples include pictures, sounds, places (i.e. on walks), metaphors, analogies and more.
The second concept is generation. This ties together with active retrieval. When you test yourself actively producing (generating) knowledge, that is more helpful than just passively re-reading it: The ability to output > the ability for input. Experimental learning is a form of generation, so is testing, summarizing or re-writing (not reading!).
Then there is reflection. A good way to think about your brain, that was prominently introduced by Kahnemann & Tversky, is to separate two types of thinking: System one is fast, almost like intuition; Imagine answering a question with the first thing that comes up, not putting in any "resistance" at all. System two is this resistance: It is more elaborate thinking, an internal dialogue going back and forth. When we are wrong about the world, it is often because of an error in system one. Reflection, a form of system two, can help us become more aware of these structural errors. It is a form of metacognition: Thinking about (the way your are) thinking.
A good help in finding these errors is calibration: We can use external tools to test our idea of something against reality. This is essentially what testing already does. The calibration part is to use the feedback from the test to maek changes in our way of learning. If you find out that you are often missing one certain part, then it makes sense to work on it more thoroughly.
A great tool for remembering large amounts of data are memonic devices. The idea is to create strong visualizations that you identify with the information you need to remember. It might go something like this: In your mind, you walk step by step through your house. You picture everything as vividly as possible. Now you create certain stops. And for each one, you identify it with the thing you need to remember. You can add objects that help you do so, and the visuals should be as vivid and intensive as possible: So much that you can not not notice them.
Related
There are many good books out there on the science of learning. I liked Moonwalking with Einstein because it is more story-like than textbook-like. There is Deep Work, Peak by Anders Ericsson, Ultralearning by Scott H Young, and many more. I don't think any of them are strictly necessary, but each one adds a new perspective. How much you like a book is not about how good the book is, but about how good it fits into your life at the time you read it. I have also observed, that I have a better ranking of books that have opened up a new topic and set of books for me. For example, I read Deep Work before all the other learning books, and thus it has a special place in my "internal library".
Another, more wholesome and out-of-the-box book is The Art of Learning by Josh Waitzkin.
And for a condensed view of the science of learning, I would recommend following Justin Skycak.