Topic: Policy Mirror Descent

Emotions significantly shape behavior, with mood---a prolonged emotional state---affecting decision-making processes by biasing perceptions of future outcomes. Recent studies suggest that mood functions as an extrapolation mechanism, tracking deviations between recent experiences and expectations to predict and inform future behavior. Understanding this mechanism has important implications for both biological systems and artificial agents. In this study, we incorporate a mood-like mechanism, modeled as momentum, into the policy dynamics of a Reinforcement Learning (RL) agent. Specifically, we