ANNECS: a neural network compiler and simulator
O.R. Vellacott · 1991
Summary form only given, as follows. ANNECS is a software tool that compiles a high-level, object-oriented specification to a functionally equivalent neural network. It does this by realizing each object in the specification as a functionally equivalent cluster of neurons and synapses. All objects are defined ultimately in terms of just two primitives, the neuron and the synapse. Thus, the compilation method consists of recursively expanding each object into its constituent objects, until the definition consists of neurons and synapses only. Since clusters of neurons are embodiments of objects whose function is fully described within the specification, the functioning of the network may be completely understood. Moreover, since networks compiled in this way are functionally equivalent to their algorithmic specification, computation theory may be applied to these neural networks. An application which demonstrates these principles is discussed, i.e. a simple robot controller which picks up objects and drops them into holes as it moves around in a world containing stairs.>