Notes on Temporal Credit Assignment

The above diagram represents a simple, obvious mechanism to solve at least some temporal credit assignment problems. It shows a process by which a conditioned stimulus that occurs sufficiently prior to the occurrence of the unconditioned stimulus that it wouldn’t ordinarily be within the period of eligibility that would produce an increase in synaptic efficacy. Suppose we postulate that the CS fires 4 seconds prior to the UCS and also stimulates a sequence of neurons that fire 1 second after the preceding one as illustrated above. If we assume an eligibility period of one second, the CS-UCR synapse would not be eligible but the +3 neuron would in fact fire just before the UCS with the result that there would be an increase in synaptic efficacy between the +3 neuron and the UCR. With a few repetitions, we would expect that, in the future, three seconds after the CS is presented, the UCR would be elicited.

The same mechanism may apply in cases of operant conditioning where the CS is replaced by the discriminative stimulus and the UCR is replaced by the command motor neuron that produces an operant behavior.