LEARNING by normative feedback
| Collection | International Encyclopedia of Systems and Cybernetics |
|---|---|
| Year | 2004 |
| Vol. (num.) | 2(1) |
| ID | ◀ 1878 ▶ |
| Object type | General information |
In a net of biological neurons, the global capacity of the net results from the self-organization of neural connections under the effects of the inputs that they receive and transmit.
L. PERSONNAZ et al. state: “The principle of this self-organization is simple: The efficiency of any synapse is reinforced when the neurons that it connects have a tendency to be active or inactive simultaneously; in the opposite case it lessens” (1988, p.1364).
This effect is strengthened locally within the strongly connected net, by repeated feedbacks.
Learning seems to be the progressive dynamic stabilization of frames of reference within the more or less narrow limits of a repertoire of stored information, relevant within a kind of pre-imprinted but very general program, while the irrelevant data are not retained.
See for example N. S. CLAYTON about song learning in birds (1991, p.466-72).
It could be a progressive algorithmization of behaviors. The upbuilding of such algorithms could possibly be modelized through attractors.
These concepts are also relevant for the new forms of perceptrons (recognition of more or less well written or printed types, for example).