Editing SuperMemo (section)

== Algorithms ==
The specific algorithms SuperMemo uses have been published, and re-implemented in other programs.

Different algorithms have been used; SM-0 refers to the original (non-computer-based) algorithm, while SM-2 refers to the original computer-based algorithm released in 1987 (used in SuperMemo versions 1.0 through 3.0, referred to as SM-2 because SuperMemo version 2 was the most popular of these).<ref>{{cite web|url=https://www.supermemo.com/en/blog/account-of-research-leading-to-the-supermemo-method-3|title=Account of research leading to the SuperMemo method|author=P. A. Woźniak|author-link=Piotr Woźniak (researcher)|date=1990|access-date=2020-11-18}}</ref><ref name="Wozniak1990-SM2">{{cite web|url=https://www.supermemo.com/en/archives1990-2015/english/ol/sm2|author=P. A. Woźniak|author-link=Piotr Woźniak (researcher)|date=1990|access-date=2020-11-18|title=Application of a computer to improve the results obtained in working with the SuperMemo method}}</ref> Subsequent versions of the software have claimed to further optimize the algorithm.

[[Piotr Woźniak (researcher)|Piotr Woźniak]], the developer of SuperMemo algorithms, released the description for SM-5 in a paper titled ''Optimization of repetition spacing in the practice of learning.'' Little detail is specified in the algorithms released later than that.

In 1995, SM-8, which capitalized on data collected by users of SuperMemo 6 and SuperMemo 7 and added a number of improvements that strengthened the theoretical validity of the function of optimum intervals and made it possible to accelerate its adaptation, was introduced in SuperMemo 8.<ref name="help.supermemo.org">{{Cite web|url=https://help.supermemo.org/wiki/SuperMemo_Algorithm|title=SuperMemo Algorithm - SuperMemo Help|website=help.supermemo.org|access-date=2019-05-01}}</ref>

In 2002, SM-11, the first SuperMemo algorithm that was resistant to interference from the delay or advancement of repetitions was introduced in SuperMemo 11 (aka SuperMemo 2002). In 2005, SM-11 was tweaked to introduce boundaries on A and B parameters computed from the Grade vs. Forgetting Index data.<ref name="help.supermemo.org"/>

In 2011, SM-15, which notably eliminated two weaknesses of SM-11 that would show up in heavily overloaded collections with very large item delays, was introduced in Supermemo 15.<ref name="help.supermemo.org"/>

In 2016, SM-17, the first version of the algorithm to incorporate the two component model of memory, was introduced in SuperMemo 17.<ref>{{Cite web|url=https://supermemo.guru/wiki/Algorithm_SM-17#Introduction|title=Algorithm SM-17|website=supermemo.guru|access-date=2019-05-01}}</ref>

The latest version of the SuperMemo algorithm is SM-18, released in 2019.<ref>{{Cite web|url=https://supermemo.guru/wiki/Algorithm_SM-18|title=Algorithm SM-18|website=supermemo.guru|access-date=2020-05-09}}</ref>

=== Description of SM-2 algorithm ===

The first computer-based SuperMemo algorithm (SM-2)<ref name="Wozniak1990-SM2"/> tracks three properties for each card being studied:

* The repetition number '''n''', which is the number of times the card has been successfully recalled (meaning it was given a grade ≥ 3) in a row since the last time it was not.
* The easiness factor '''EF''', which loosely indicates how "easy" the card is (more precisely, it determines how quickly the inter-repetition interval grows). The initial value of '''EF''' is 2.5.
* The inter-repetition interval '''I''', which is the length of time (in days) SuperMemo will wait after the previous review before asking the user to review the card again.

Every time the user starts a review session, SuperMemo provides the user with the cards whose last review occurred at least '''I''' days ago. For each review, the user tries to recall the information and (after being shown the correct answer) specifies a grade '''q''' (from 0 to 5) indicating a self-evaluation the quality of their response, with each grade having the following meaning:

* 0: "Total blackout", complete failure to recall the information.
* 1: Incorrect response, but upon seeing the correct answer it felt familiar.
* 2: Incorrect response, but upon seeing the correct answer it seemed easy to remember.
* 3: Correct response, but required significant effort to recall.
* 4: Correct response, after some hesitation.
* 5: Correct response with perfect recall.

The following algorithm<ref>{{cite web |title=Super-Memo 2 Plugin for Super-Memo for Windows: Delphi Source Code |url=https://www.super-memory.com/english/ol/sm2source.htm |website=SuperMemo Articles |access-date=23 August 2021}}</ref> is then applied to update the three variables associated with the card:

 '''algorithm''' SM-2 '''is'''
     '''input:'''  user grade ''q''
             repetition number ''n''
             easiness factor ''EF''
             interval ''I''
     '''output:''' updated values of ''n'', ''EF'', and ''I''
 
     '''if''' ''q'' ≥ 3 ''(correct response)'' '''then'''
         '''if''' ''n'' = 0 '''then'''
             ''I'' ← 1
         '''else if''' ''n'' = 1 '''then'''
             ''I'' ← 6
         '''else'''
             ''I'' ← round(''I'' × ''EF'')
         '''end if'''
         increment ''n''
     '''else''' ''(incorrect response)''
         ''n'' ← 0
         ''I'' ← 1
     '''end if'''
     
     ''EF'' ← ''EF'' + (0.1 &minus; (5 &minus; ''q'') × (0.08 + (5 &minus; ''q'') × 0.02))
     '''if''' ''EF'' < 1.3 '''then'''
         ''EF'' ← 1.3
     '''end if'''
     
     '''return''' (''n'', ''EF'', ''I'')

After all scheduled reviews are complete, SuperMemo asks the user to re-review any cards they marked with a grade less than 4 repeatedly until they give a grade ≥ 4.

=== Non-SuperMemo implementations ===
Some of the algorithms have been re-implemented in other, often [[free software|free]] programs such as [[Anki (software)|Anki]], [[Mnemosyne (software)|Mnemosyne]], and [[Org-mode|Emacs Org-mode]]'s Org-drill. See full [[list of flashcard software]].
 
<!-- Please don't link to anything which doesn't have a WP article. These are just the most notable clones; this isn't meant to be a comprehensive link-dump paragraph. -->
The SM-2 algorithm has proven most popular in other applications, and is used (in modified form) in Anki and Mnemosyne, among others. Org-drill implements SM-5 by default, and optionally other algorithms such as SM-2 and a simplified SM-8.