A functional neural network prototype for multidimensional data analysis

Artificial neural networks present a powerful tool to analyse complex systems. They have been long used to tackle the difficulties of image analysis and interpretation with applications ranging from character recognition (Howells et al. 1996) to colour image processing (Howells et al. 1997). This paper presents a novel approach which exploits the generality of expressions available in the area of constructive type theory and its potential for the production of guaranteed bug free, provably correct software to develop a working neural network prototype. To illustrate the possible applications of our work we present results from two examples arising from co-operation with the Dover Harbour Board.