Trade.Information

Details of What learns a computer

The main mechanism of learning of a computer is based on the process of memorization of the information. The type of codifies door to states of the R-tutto.o.niente type, but using one logical fuzzy it can memorizzare information of "vanished" type that they concur to manage variable subject to uncertainty and incomplete acquaintance.

L`introduzione of algorithms of chaotic type concurs to render the system typically unforeseeable supplying that human characteristic of the contingency and irripetibilità of the answers. The computazionali rules, with which the logons between the data that constitute the memory they are elaborates to you, could be strengthened or be weakened so as to to improve the total performances second criteria determine dall`ambiente outside to you.

E` possible to use analogous techniques to those typical ones of the genetic algorithms, in which you e` one natural selection between different algorithmic procedures: l`algoritmo more effective it is born for a the most adapted species of survival of.

In adding you can be accidental elements incorporates to you in the way in which the machine it must modify the own performances. E` in particular possible to use the variable thunderstorm in order to obtain, through chaotic processes, an effect of pure chance. L`apprendimento can be controlled from an expert human or be based on coherence algorithms and not contradiction. In particular two types of input can exist: that artificial and that natural one.

As far as l`input artificial images teaching that the truth communicates to the mathematical machine and tries to guide it towards l`ottenimento of a way in order to distinguish the truth from the falsità. L`insegnante can inform the machine when it has made an error or communicate them varies concepts matemati to us and different acceptable methods of mathematical demonstration.

As far as l`input natural physical dall`ambiente it could be had "mutuate ideas" from the behavior of the physical objects. L`ambiente could also offer concrete realizations of concepts matemati to us.

However with passing of the time teaching and the l`ambiente physicist could less and less be essential for the formation of the judgments of the machine, in particular would become inner the computazionale power more and more important.

The criterion contradditorietà would not become more and more predominant. If the machine must have really the abilities to a human being, will need one species of concept of "inconfutabile mathematical truth". Hour the concept of "inconfutabile truth" of a human mathematician cannot be obtained from whichever with of mechanical rules that are conoscibile and completely credible.

If we suppose that a machine is able to catch up the level of capacita` mathematical of a human mathematician, then also its concept of inconfutabile mathematical truth must be something that cannot be achieved from whichever with of mechanical rules perceived like valid.

This point brings to light the difficulties intriseche in the realization of a learning algorithm that can be of general and at the same time efficient character.