An artificial neural network study of the relationship between arousal, task difficulty and learning

We compare the performance of a backpropagation neural network and a recirculation neural network when they are used to simulate the interactions between arousal, task difficulty and learning. We use number strings that vary in terms of their randomness as the stimuli to be learned and the bias unit to simulate arousal. We find that the recirculation neural network shows an interaction between task difficulty and arousal which is typical of that observed when living organism learn a task. This interaction is not observed with the backpropagation algorithm. We conclude that the recirculation neural network provides a better model of arousal and learning than the backpropagation algorithm.