A functional framework for the specification of artificial neural networks

The authors consider artificial neural networks (ANNs) as first class objects with a double identity: a functional one, when the network is used in activation mode (the input units are clamped, the network evolves, and the results are observed on the output units, possibly the same), and a structural one, where one is interested in the internal states of the neural network. This duplicity is easily expressed in functional languages like Lisp, and besides its elegance it has interesting properties such as network composition and easy integration within symbolic reasoning. The ultimate objective of this research project is to develop an environment in which the definition and the use of ANN and the traditional symbolic processing are strongly coupled