NEUCOMP - A NEURAL-NETWORK COMPILER

NEUCOMP is a Neural Network compiler that compiles the program of a particular NN model written as a list of mathematical specifications (known as NEUCOMP language) and translates it into a chosen target program. The mathematical specifications used are represented by scalar, vector and matrix assignments. The NEUCOMP language is a procedual language for general purpose NN models. It combines with an existing graphical package which can portray the NN architecture and display a graph of the results. The NN models being considered so far are the Multi-layered Perceptron, Kohonen Self-Organizing Network and Adaptive Resonance Theory (ART). The Multi-layered networks use a supervised learning algorithm whilst the Kohonen and ART networks use unsupervised learning.