Fast forward to the next semester, and I'm taking an introductory class in neural networks. This class, as you can imagine, requires some of the material introduced in those earlier classes.
But even though barely 3 months have passed, I struggled to remember how to do basic operations and had to relearn simple concepts. This was especially problematic while trying to understand backpropagation, since I constantly had to interrupt myself because I had to look something up again. But in the end, I managed to get it by sinking more time into it.
Fast forward to today, and I failed to remember how backpropagation works while taking another class about machine learning. And I'm not just talking about details, I can't even remember some parts of how it works abstractly.
This is extremely frustrating, because it feels like I'm wasting my time by learning something, only to forget it within a few months.
Is there anyone who struggled with the same thing and found some way to alleviate this issue?
When it comes to programming languages, for instance, the first thing I teach myself is how to look up answers to questions like "how do I look up the length of a string?" in the official documentation. This is much more reliable than using Google or Bing to look up frequently wrong answers on Stackoverflow and programming splogs. This works really well for Java and Python, but not so well for Clojure (there is a book you've got to buy) and is also tricky for Javascript (MDN's documentation is close to an official manual)
Each time you "learn" something and incorporate it into your docs, and then work with it, it becomes a little more familiar and after a long time you really will master the material.
Another "hack" is to learn how to derive things yourself. For instance the other day I had to take some derivatives a little more complex than what I usually do and it's been 30 years since I took calculus, I remember some of the rules but I forgot some that I need and just took a few minutes to derive them from first principles.
In the case of something like the Fast Fourier Transform or backpropagation you should be able to derive the method from scratch knowing just a few facts about how it works.
I really identify with this feeling. For a time, this meant that I put the minimum effort into studies to get decent grades and completely stopped learning on my own. Thankfully, soon enough I found Anki [0] and everything changed.
Through using the app ~daily for over 7 years now, I have a much better idea of how my memory works. It doesn't mean that I never forget anything (even things in Anki will decay if not engaged with), but it makes memory more of a choice rather than something that will automatically aggressively fade.
I'd recommend trying it out for 2-3 weeks and seeing if you like it. Using a shared deck that someone has already put together for something that you're already interested in is a great way to start. For me, that was geography [1].
It's also been said, "You have to know it to write it or teach it" [1]
If you're not staying in the material between classes, you'll have to "relearn" at least parts of it
I got great grades in financial accounting 2+ decades ago
Doesn't mean I can do financial accounting today (heck! Probably couldn't "do" it even a semester later)
I did fantastically in an intro to COBOL class in 2000. Haven't so much as glanced as COBOL since (to the best of my knowledge)
You'd got to keep using what you learn, or it goes stale/disappears
Your memory's a muscle [2] - you've got to give it a workout, or it will atrophy [3]
--------
[0] https://www.flintrehab.com/use-it-or-lose-it etc
[1] https://www.electronicdesign.com/technologies/dsps/article/2...
[2] https://www.dummies.com/article/body-mind-spirit/physical-he...
- Take B. monnieri. It's not a huge effect but it does seem to have a real positive effect on memory if you take it for a while.
- Practice n-back
- Get good sleep and exercise frequently to optimize your brain health
Try implementing a basic neural network and backprop in something like numpy. If you really want to go hardcore, implement it again without numpy.
https://www.qwant.com/?q=improve+memory+site%3Anews.ycombina...
it doesnt make sense to beat yourself up for not remembering everything
one good perspective is you may forget everything about it, but now you still know how to find the answers that lead you to it.
- Nothing can really significantly improve your raw baseline memory ability
- Familiarity with a domain improves your brain's ability to compress the information and memorize more/faster in short-term memory, but not for longer.
- Memorization strategies are powerful. Spaced repetition[1] is probably the most powerful tool known to mankind, since it works around our natural propensity to forget information that doesn't see regular use. Possibly only surpassed by incremental reading/writing. Memory palaces also actually work, and probably could be combined with it.
[1] https://senrigan.io/blog/chasing-10x-leveraging-a-poor-memor...