MAP-BASED GENERALIZATION AND ITS AFFECTIVE MODULATION IN A DYNAMIC REWARD ENVIRONMENT

In dynamic environments, decision-makers must balance strategies that differ in flexibility and cognitive cost. One approach incrementally learns stimulus–reward associations through trial and error, while a more adaptive strategy exploits structured knowledge of the environment to infer rewards via cognitive maps. While humans are known to use both strategies in simple tasks, it remains unclear how they balance these strategies when reward structures change rapidly but in accordance with an underlying relational structure, and how affective salience may bias this balance. To address these questions, we developed a novel experimental paradigm that dissociates map-based generalization from model-free learning. On day 1, participants learned the relational structure among stimuli. On day 2, they made repeated choices in an environment where rewards changed frequently but followed the learned structure. Behavior was best captured by a hybrid model combining map-based inference and model-free learning.
In a follow-up study, a conditioning block was introduced where one stimulus was paired with an unpredictable aversive scream immediately after map acquisition. We examined whether this manipulation reshapes the strength of map-based inference during subsequent decision-making on day 2, consistent with the hypothesis that fear enhances the strength of relevant structural representations, especially in anxious participants.
Together, these studies provide insights into how individuals arbitrate between map-based generalization and stimulus-outcome learning in dynamic and affectively charged environments, with implications for maladaptive decision-making in psychopathology.