FUNCTIONAL SPECIFICATION OF A NEURAL NETWORK**This research was partly sponsored by the Dutch Neural Network Foundation (original) (raw)

Artificial Neural Networks, 1991

Abstract

Functional programming languages have been successfully employed to make executable neural network descriptions. For the sake of efficiency, the programs are mixtures between descriptions of the behaviour and the data-flow needed to obtain an efficient algorithm. This has a bad influence on the descriptions' readability. In this paper, we will present a functional description of a neural network which isn't hampered by any efficiency considerations. Our description of a neural network is a mathematical specification of that network. Using an ordinary functional programming language evaluation strategy, our network description has a time complexity which is exponential in the number of learning steps of the network. Using an evaluation strategy which is improved with a memoizing facility, the specification has an almost linear time complexity in the number of learning steps.

Rinus Plasmeijer hasn't uploaded this paper.

Let Rinus know you want this paper to be uploaded.

Ask for this paper to be uploaded.