Increasing the biological inspiration of neural networks

Abstract

Starting from a nerve cell functional characterization, we define formally the autonomous learning to concatenate sequences and prove it to be a possible solution for the problem that faces the, eg vertebrate, nervous systems: ie, to choose and to store, without outside help, the instructions to compute the actual sensor/effector correspondences they have to control. In our formal system we assign the initial connection matrix elements so that the rules, namely the Caianiello relation iterated application, autonomously and deterministically control the meta-rule, namely the Hebbian rule, application. © 2002 Springer-Verlag Berlin Heidelberg.