Date: 18 Jun 2005

A generic formulation of neural nets as a model of parallel and self-programming computation

* Final gross prices may vary according to local VAT.

Get Access

Abstract

In the same way the more conventional fields of computer science need some theory, including the mathematical foundations of the calculus and the establishment of formal models, neural computation also needs its own. The basic requirements of this model are modularity, “small grain”, high connectivity, parametric local computation and some capacity of self-programming by means of the adjustment of these parameters.

We present here a proposal in this line that allows the integration in a single frame of all current models (analogic, logic and inferential) and makes clear the natural way to bridge the symbolic and connectionistic perspectives of AI extending the model of local computation to hierarchic graphs, building networks by joining graphs and studying the set of operators we need for modifying local computation parameters values.