The order of the presentation of the samples can lead to different vectors. Figure 33 shows the modification of the vector weight W by presenting the sample then (A) and the modification occurred by presenting then (B). Results are different.

Various modes of initialization can be assumed like a random choice, or a uniform distribution.

The learning strategy may be based on a very significant number of neurons, then the neurons whose weights do not change or change in a chaotic way are isolated. The weights of all the neurons may also be adjusted proportionally to their answer. If information on the classes are known, the process can be sped up.