23 Sept 2026: Reinforcement learning states and the dynamics of mood

Why do our moods change over days, sometimes without an obvious change in our circumstances? Computational accounts propose that mood reflects both rewarding and punishing experiences and how much we learn from them. I will present evidence that fluctuating reinforcement learning states help explain the emergence of positive and negative moods in everyday life.

In a 28-day study, 318 participants completed smartphone-based learning tasks twice daily and reported their mood and daily experiences four times daily, alongside wearable EEG and ECG recordings. A Bayesian reinforcement learning model used choices between stimuli learned on different days to identify dissociable changes in reward and punishment learning. States of amplified reward learning predicted subsequent increases in positive mood, whereas amplified punishment learning predicted increases in negative mood, across lags extending to seven days and accounting for prior mood. These learning states were also reflected in cortical and cardiac responses to reinforcement and moderated the relationship between everyday experiences and subsequent mood. Stronger coupling between punishment learning and later negative mood was associated with greater trait negative emotionality.

These findings support a computational account in which changing learning states influence how experiences accumulate into sustained affective states. I will discuss how connecting learning, physiology, and mood across naturalistic timescales can clarify mechanisms of psychiatric vulnerability and motivate new approaches to studying mood disturbances.