This paper considers neural computing models for information processing in
terms of collections of subnetwork modules. Two approaches to generating su
ch networks are studied. The first approach includes networks with function
ally independent subnetworks,, where each subnetwork is designed to have sp
ecific functions, communication, and adaptation characteristics. The second
a,approach is based an algorithms that cart actually generate network and
subnetwork, topologies, connections, and weights to satisfy specific constr
aints. Associated algorithms to attain these goals include evolutionary com
putation and self-organizing maps. We argue that this modular approach to n
eural computing is more in line with the neurophysiology of the vertebrate
cerebral ail colter, particularly with respect to sensation and perception.
We also argue that this approach? has the potential To aid in solutions to
large-scale network computational problems-an identified weakness of simpl
y defined artificial neural networks.