AUTONET is a system for the specification and simulation of neural networks, It deals with the network as a directed graph, the nodes of which are automata which accept a tree of inputs and produce a single output, and the lines of which are queues of arbitrary length via which the cellular automata communicate.

The simulation system includes the WARP network construction language. The important features of WARP are its ability to manipulate not only single cells, but also arbitrary groups of cells, as distinct units. Projection of one cell group upon another is facilitated by this feature, and also by the ability to permute the interconnections between cells, This permits convergent and divergent projections in many dimensions to be handled with ease.

The experiments conducted with AUTONET have been largely with simple and moderately complex networks. Some apparently trivial networks are shown to actually possess a wide variety of possible rhythmic outputs. These output patterns are characterized as to their stability, and to the connectivity patterns which make the various patterns more OF less likely to be in a network subject to random effects. generated

Considerable simulation of the gastric and pyloric subsystems of the lobster stomatogastric ganglion was also done, It is shown that all of the rhythms generated by the ganglion may be explained by decomposing the networks into groups of two or three cells, which smaller networks have already been well-characterized.

Evidence is also presented for the hypothesis that the endogenous bursting cells of the pyloric subsystem are not the generators of the pyloric rhythm, Rather, their purpose seems to be as synchronizers of the cycle.