Theoretical foundations of SuperMemo

The optimization technique used in SuperMemo is based on the work of a Polish biologist, Piotr A.Wozniak, who has developed a mathematical model of the decline of memory traces. The model makes it possible to determine the optimum spacing of repetitions for any desired level of knowledge retention.

SuperMemo uses Wozniak's model, and can be programmed to produce knowledge retention in the 90-99% range.

Although the details of the optimization technique are covered by SuperMemo World's trade secret, the general outline of the used algorithm is presented below.

All forms of learning produce molecular changes in neuronal synapses which form connections between nerve cells in the nervous system. These changes are gradually obliterated in the process of forgetting, which plays an important evolutionary role in optimizing the organism's responses to the outside environment.

Forgetting affects all synapses that are able to learn, and can be prevented only by means of repetition. Every attentive learner knows, that forgetting can ruin the delicate fabric of knowledge that may take months or years to build. Latin saying Repetitio mater studiorum est is as old as the art of learning.

However, not everybody knows that it is not possible to learn something once and to remember it forever without repetition!

Even a person's own name could be forgotten if not used, or in other words, repeated as often as it is. The problem with repetitions is that they consume one of the most valuable assets in the modern lifestyle, time. Therefore, the key to effective learning is to find ways to reduce the number of repetitions needed.

SuperMemo allows learning to approach the maximum natural capacity of the human brain to form memories. This is done by the optimization of repetition scheduling, i.e., finding out when and which portions of knowledge should be repeated. Given a piece of knowledge, two criteria are used to determine the length of optimal intervals which should separate repetitions: (1) minimization of the number of repetitions, and (2) maximization of overall knowledge retention.

In other words:

Because forgetting has a stochastic nature, i.e., it cannot easily be predicted when a given piece of knowledge will be forgotten, a statistical approach to the process of learning has to be applied. By statistical analysis, we can determine when a given proportion of memorized knowledge will be forgotten; hence, the following definition of optimal intervals: optimal intervals are intervals that result in a small, previously determined fraction of knowledge being forgotten. This fraction, called the forgetting index, may be chosen by the learner, and usually falls in the range 5 to 15 percent. By using optimal intervals in the process of learning, SuperMemo produces an incredible increase in the rate of knowledge acquisition without affecting knowledge retention. Optimal intervals will vary for different sorts of knowledge and different learners. The former problem is dealt with in SuperMemo by splitting the learned knowledge into the smallest possible pieces called items.

The optimization procedure, i.e., computation of optimal intervals, is then applied to each of the pieces separately, producing a unique repetition spacing in each of the cases. The principle of applying items of maximum simplicity is called the minimum information principle.

The problem of differences between learners is solved in SuperMemo by application of self-modifying algorithms that adjust repetition spacing to individual needs.

SuperMemo procedures can detect what sort of learning and what sort of learner are subject to optimization.

The net result is that a determined level of knowledge retention may be maintained in the process of learning to approach the maximum natural speed at which the learner's brain can form memories.

See also Outline of the SuperMemo optimization algorithm