A Theoretical Neuronal Learning Mechanism That Predicts the Basic Categories of Classical Conditioning Phenomena

Abstract

1. THE NEURONAL MODEL AND LEARNING MECHANISM 21 It is suggested that the neuronal model proposed by Hebb (1949) be modified in the following ways to make it consistent with the animal learning phenomena it is intended to explain: (1) instead of correlating pre-and postsynaptic levels of activity, changes in pre-and postsynaptic levels of activity should be correlated; (2) instead of correlating approximately simultaneous pre-and postsynaptic signals, earlier presynaptic signals should be correlated with later postsynaptic signals. More precisely and consistent with the first modification, earlier changes in presynaptic signals should be correlated with later changes in postsynaptic signals. Thus, sequentiality replaces simultaneity in the model. (3) A change in the efficacy of a synapse should be proportional to the current efficacy of the synapse, accounting for the initial positive acceleration in the classic S-shaped acquisition curves observed in animal learning. The resulting neuronal model is an extension of the Sutton-Barto (1981) model, which, in tum, may be seen as a temporally refined extension of the Rescorla-Wagner (1972) model. The model to be proposed is termed a drive-reinforcement model because it suggests that nervous system activity can be understood in terms of two classes of signals: drives, which are defined to be signal levels, and reinforcers, which are defined to be changes in signal levels. Mathematically, the learning mechanism of the drive reinforcement model may be characterized as follows: Aw;(t) = Ay(t) Lcij W;(t-j)jAx;(t-j) j=l (1) where w;(t) is the weight or efficacy of synapse i at discrete time, t; Aw;(t) is the change in efficacy of synapse i at time, t, and is equal to w;(t + 1)-w;(t); T is the longest A.