The Regret Motivated Reinforcement Learning

Abstract

The reinforcement learning model has been shown to explain the decision-making process of humans and animals accurately. However, standard reinforcement learning models do not contain the influence of emotion, which should contribute to humans' and animals' decision-making. This paper focuses on “Regret,” which we defined mathematically as a term with a maximum reward minus a current reward. This paper shows how regret motivates reinforcement learning and the potential application to the study of problem gambling.