On a Learning Neural Network

In this paper we present a modified version of a McClloch and Pitts network togheter with its associated language. We also give the assembler needed to translate a program’s symbolic form, viz. any wff belonging to the prepositional calculus, into the values associated to the couplings between the network nodes, namely into the “coupling coefficients” values. As an example of the practical application for the said results it is presented the program for the sum of two binary numbers together with the corresponding coupling coefficients values. It is then discussed the role that the memory can play as a metarule for the given language and a way to store into the coupling coefficients values the relevant interactions between the network, seen as a data driven machine, and an unpredictably evolving environment. Since to change the coupling coefficients alters the stored programs, viz. the network responses to the environment stimuli, this is a way to implement learning.