C. Ferretti, G. Mauri
NNET: some tools for neural networks simulation.
Abstract
In this paper we describe the development and use of a modular system for
Neural Network simulation, called NNET. Our aim was to provide some tools both
flexible in use and efficient in execution. Things to be simulated are
collections of many simple processing elements, the neurons, connected each
other, potentially running in parallel and communicationg through the
connections between them. The implementation of this computational structure
requires efficient hardware to reproduce parallel processing and good software
to permit easy definition and testing of the (often complex) structure of a
net of connected neurons. NNET is based on a neural networks simulating
nucelus, with a library of efficient 'C' functions describing some current
model, a compiled low level language for the description of nets, a text-video
interface to assist the user during the simulations, and an interpreted
structured language to control the operations on all the simulating
environment. Some possible further developments of NNET are discussed.