Effect of incomplete memorization in a computational model of human cognition

Abstract

Dropout has been introduced as a simple yet effective method to prevent over-learning in deep learning. Although its mechanism, i.e., incapable of utilizing all memorized units, seems quite natural to human cognition, the effect of dropout on models of human cognition has not been addressed. In the present research, we apply dropout to a computational model of human category learning. We compared models with and without complete memorization abilities, and results showed that they differed acquired association weights, dimensional attention strengths, and how they handled exceptional exemplars.