Algorithm SM-18

This text is part of: "History of spaced repetition" by Piotr Wozniak (June 2018)

General idea

Algorithm SM-18 is the newest version of spaced repetition algorithm used in SuperMemo (2019). The main difference is the new model for estimating item difficulty. It is a departure from the original assumption that item difficulty is constant. We have clear evidence of dramatic changes in item difficulty in the course of learning. Those changes can be explained by anchoring, when new mnemonic context converts a difficult item into an easy item literally overnight. Anchoring can occur without any changes to the contents of the item. Naturally, due to interference, the reverse process may also occur (i.e. items may suddenly become difficult). However, forgetting was handled efficiently in Algorithm SM-17.

As of May 2019, Algorithm SM-18 is in use in SuperMemo 18. Once it passes all necessary benchmarks, it will be used in all SuperMemo products.

Explanation for babies

The details of Algorithm SM-18 are not much different from Algorithm SM-17. However, we often hear a complaint that the explanations are too convoluted. Users want to know more or less how the algorithm works. This is the baby version:

1. compute startup interval using the first forgetting curve
2. at the end of the interval, make a repetition
3. after a repetition, compute new difficulty
4. estimate retrievability using three sources of information (see: Algorithm SM-17)
5. estimate stability using three sources of information (see: Algorithm SM-17)
6. compute the next optimum interval using the stabilization curve
7. update the memory matrices used in computations: stabilization matrix (SInc[]), recall matrix (Recall[]), etc.
8. go back to Point 2 (i.e. next repetition)

Details of the algorithm

The differences between Algorithm SM-17 and Algorithm SM-18 are not significant except for the way the difficulty is computed. This is why you can learn more details by reading about Algorithm SM-17 (2016).

This text is part of: "History of spaced repetition" by Piotr Wozniak (June 2018)