zlacker

Isn't FSRS (the new algorithm used in Anki since a few years ago) already based on machine learning?

>>flexag+(OP)
Yes. Stochastic gradient descent, to be precise.

https://github.com/open-spaced-repetition/fsrs4anki/wiki/The...

>>flexag+(OP)
Too old school and too effective.

FSRS just works, even without a GPU so it's not the cool kind of AI / machine learning these days.

No joke though: the FSRS model is marvelous, and Anki remains one of the best free + open source implementations around.

I've been learning German recently and Anki (in FSRS mode) is one of the most important learning tools I have. No joke.

------

Every card remembers every rating you give it, as well as the time / date. This allows for Anki to solve for a 'forgetting curve', and predict when different cards have a chance to be forgotten.

There is furthermore the machine learning / stochastic descent algorithm to better fit the assumed forgetting curves to your historical performance. This is the FSRS Optimize parameters button in the settings panel.

>>dragon+Iq
> Every card remembers every rating you give it, as well as the time / date. This allows for Anki to solve for a 'forgetting curve', and predict when different cards have a chance to be forgotten.

True to a point; every card has its ratings, but the "forgetting curve" algo of FSRS is only tuned to the deck (or "option set") that the card is in, not per card.

>>michae+cz4
The entire FSRS parameter set (~20+ parameters, depending on FSRS version) is per deck.

Each card is tuned to... 2 parameters IIRC? f(Difficulty, Stability, Time) == Retrievability. Time is just time so its not really a parameter, but Difficulty and Stability is solved on a per-card basis.