Spaced repetition

When you have memorised something, you need to review that material, otherwise you will forget it. However, as you probably know from experience, it is much more effective to space out these revisions over the course over several days, rather than cramming all the revisions in a single session. This is what is called the spacing effect.

During the past 120 years, there has been considerable research into these aspects of human memory (by e.g. Ebbinghaus, Mace, Leitner and Wozniak). Based on the work of these people, it was shown that in order to get the best results, the intervals between revisions of the same card should gradually increase. This allows you to focus on things you still haven't mastered, while not wasting time on cards you remember very well.

It is clear that a computer program can be very valuable in assisting you in this process, by keeping track of how difficult you find a card and by doing the scheduling of the revisions. Let's see how this works in practice in the Mnemosyne program.

The software will present you with a question, and your task is trying to remember the answer. Afterwards, you rate yourself on a scale between 0 and 5. These ratings will be used in computing the optimal revision schedule. Let's see what these grades mean, both for learning new cards and for reviewing cards you've memorised earlier.

Memorising new cards

When you are learning new cards, use grades 0 and 1 to signal that you have not yet memorised them. Grade 1 cards are becoming more familiar than grade 0 cards, and will be repeated less often.

These cards will be repeatedly shown until you give them a grade 2 or higher. This means that you've memorised the card and that you'll be able to remember it for one or two days. The 'Not memorised' counter will decrease by one.

This card will be scheduled at some future date, when you're likely still be able to remember it with some effort, without having forgotten it completely. This is most efficient for the learning process.

Reviewing previously memorised cards

If you study these cards again tomorrow, the 'Scheduled' counter will tell you how many previously memorised cards you need to review. These are grade 2 to 5 cards.

If a card reappears too soon, and you're able to remember it without any effort, rate the card a 5. The interval to see this card again will be a lot longer.

If the interval is just right, so that you remember it, albeit with some effort, use grade 4.

If, however, it takes you significant effort to remember the answer, and you think the interval was too long, then rate the card 3 or even 2.

If you fail to remember it altogether, rate it either 0 or 1, and after you have finished reviewing all the scheduled cards, it will appear repeatedly until you think you'll be able to remember it again for a few days.

For best results, it is suggested to do your revisions every day, although Mnemosyne will try to cope as well as possible if you postpone your revisions or if you want to learn ahead of time.

Working with large numbers of new cards

When you are learning a large number of new cards, it does not make sense to try and learn them all at the same time. For this reason, Mnemosyne has a setting Work on at most X non-memorised cards at the same time, which defaults to 10. This means that at any given time, you will not be trying to memorise more than 10 new cards. Note that this setting does not tell you how many new cards you need to learn per day. You are the judge of that: you can learn more cards or less cards, depending on how you feel.

However, we recommend only going through a limited number of new cards each day, in order to help Mnemosyne achieve a better spread of your workload when reviewing those cards again.

I'm worried I'm doing it wrong!

Mnemosyne user Michael Campbell had this to say on the mailing list:

Having used Mnemosyne a few years now and watching this mailing list pretty closely I've seen a topic surface over and over again; and that is what happens if I do something outside the "perfect" parameters; the 2 biggest being "what if I miss days?", "oh my, I've graded a card wrong; what do I do?". The answer is, "nothing". It's no big deal. Let it go. If you graded it too HIGH, you may forget it next time, which will be taken care of by your 0 or 1 grade when you next see it. If you grade it too low, you'll see it earlier than you would have otherwise, and what's the big deal? And for missed days, just do what you can when you can. It can be shown that assuming you remember at least some of the time from day to day, you can do 1 card a day and eventually get through an arbitrarily large stack.

Ones memory is such a fluid dynamic thing that trying to curve-fit SM2 (or any algorithm) to it just isn't possible, or beneficial. Your memory may kick ass today, and absolute shite tomorrow; no algorithm can hope to model that. And each person is different too. I think these fine-tuning exercises I see people attempting, while perhaps fun, are of little to no actual benefit. The benefit comes from doing, not tweaking. I'm sure Peter or Gwern have some studies at hand that might have more information.

So, make a best effort on grading, and try to do it daily or at least as often as you can, and it'll work fine. This isn't an optimization exercise, it's just meant to reduce work that may not be necessary.