An Investigation of the Behavior of Coupled Networks of Neuromime Devices

摘要

The complexity of the biological neuron led investigators to the study of networks of simplified neuron models. It was shown that any possible device possessing a known input-output relationship can be constructed using Boolean algebra and logical calculus, although large numbers of them are required for complex behavior. The small memory capacity of the fixed logic type of neuron model, as well as other disadvantages, caused some investigators to turn attention to the adaptive neuron model. The behavior of a coupled network of 15 of these neurons is investigated by use of digital computer simulation, and the 'learning' ability of the network is studied for different neuron parameters. It is shown that the topology and the initial values of the network parameters play a vital role in the ability of the network to 'learn', and it is concluded that, for the topology chosen, the simulation of the Rochester neuron model produces a network that apparently can 'learn' an input and can reproduce it later.