|
@michael_nielsen | |||||
|
Striking analogy between the mnemonic medium and supervised learning. Makes me think about analogies to things done in supervised learning: adding noise to the data (questions) to improve generalization; changing the loss function (overall scoring), etc. Fun to think about! twitter.com/mlpowered/stat…
|
||||||
|
||||||
|
michael_nielsen
@michael_nielsen
|
16. sij |
|
In the next essay we are, in fact, doing something much like adding noise to the questions (or, at least, variation) to help improve generalization.
|
||
|
|
||
|
michael_nielsen
@michael_nielsen
|
16. sij |
|
Incidentally, a very amusing instance of this idea is sans forgetica, a typeface designed to improve retention because it's so damned difficult to read: en.wikipedia.org/wiki/Sans_forg… I have trouble believing this works, but, well, you can read the paper yourself. pic.twitter.com/Nr7R5sNfWX
|
||
|
|
||
|
José Luis Ricón (Artir)
@ArtirKel
|
16. sij |
|
An analogy I use when learning a new domain is that learning from just one place may lead to overfitting, and it's good to see the context presented in different ways, linked to different things each time
|
||
|
|
||
|
José Luis Ricón (Artir)
@ArtirKel
|
16. sij |
|
Also helps to get a sense of how robust the literature is and how much debate is there.
|
||
|
|
||
|
POLAR - Personal knowledge repository
@getpolarized
|
16. sij |
|
... and once I got the supervised learning metaphor I was addicted and that's why I started building Polar. We now have our own implementation of SM2 and sync with #anki and I think the fully integrated environment is really addictive.
|
||
|
|
||
|
Tom Bielecki
@tombielecki
|
16. sij |
|
Will Polar someday have BiDirectional Links and Indented lists/references like Roam Research? I really want to go back to Polar for SRS but Roam's workflow fits my brain better
|
||
|
|
||
|
Emaad Manzoor
@emaadmanzoor
|
17. sij |
|
You may find the Memorize project (learning.mpi-sws.org/memorize/ ) by @btabibian @musically_ut @autreche and team relevant. They have an accompanying paper in PNAS '18: "Enhancing human learning via spaced repetition optimization"
|
||
|
|
||
|
POLAR - Personal knowledge repository
@getpolarized
|
16. sij |
|
The analogy of supervised learning and spaced repetition I think is dead on. I've been referring to Polar as a "cognitive prosthetic". My background is partly in machine learning so when I got into #spacedrepetition I immediately realized it was supervised learning for humans.
|
||
|
|
||
|
Phil Mohun
@philmohun
|
16. sij |
|
What's the best tool for spaced repetition? I've tried the ANKI app, but it's clunky. Any good implementations?
|
||
|
|
||
|
Tom Bielecki
@tombielecki
|
16. sij |
|
Have you tried Quizlet? I think it has an underlying SRS algorithm, and really user friendly
|
||
|
|
||