ABSTRACT
From the late 1960s to the early 1980s, researchers in artificial intelligence largely abandoned neural networks of the linear threshold variety in favor of heuristic computer programs. A number of distinctions were found between the capabilities of perceptrons with different reinforcement rules and different training procedures, distinctions which are mainly of historical interest. Neural models frequently average random single-neuron effects across the functional groups of neurons that constitute network nodes; hence, the interactions between nodes become deterministic. Application to neural network modeling of analogies with other fields remained popular for several decades. In a large number of the neural network models developed since then the functional units are neuron populations rather than single neurons. C. L. Hull’s stimulus traces can be considered as the amounts of activity of particular nodes or functional units in a neural network.
