A reinforcement learning model of bounded optimal strategy learning

In this paper we report a reinforcement learning model of how individuals learn the value of strategies for remembering. The model learns from experience about the changing speed and accuracy of memory strategies. The reward function was sensitive to the internal information processing constraints (limited working memory capacity) of the participants. In addition, because the value of strategies for remembering changed with practice, experience was discounted according to a recency-weighted function. The model was used to generate predictions of the behavioural data of 40 participants who were asked to copy appointment information from an email message to a calendar. The experience discounting parameter for a model of each individual participant was set so as to maximize the expected rewards for that participant. The predictions of this bounded optimal control model were compared with the observed data. The result suggests that people may be able to choose remembering strategies on the basis of optimally discounted past experience.