In this kind of learning, the samples are known, and therefore the number of classes to be distinguished. With a configuration of known input (sample) the answer to be found is given. If the answer is not expected, the vectors of weight are modified. The process runs with an iterative behaviour onto the set of samples until the stabilization of the weights is reached.

According to the situation and especially the number of classes, the order of presentation of the samples is important. All the samples of a class are presented, then all the samples of another class and so on. This can lead to less recognize the sample of the first classes, that expresses a kind of amnesia.

The information of the samples can also be alternated to take this problem into account: a sample by class and so on, but this can lead to badly distinguish the classes between them. This shows the problem of the "pedagogy" to be used.