The precision of attention selection during reward learning influences the mechanisms of value-driven attention

Reward-predictive items capture attention even when they are irrelevant to current goal. While previous studies suggest that value-driven attention generalizes to items sharing critical reward-associated features (e.g., red), recent findings propose an alternative generalization mechanism based on context-dependent feature relationships (e.g., redder). Here, we examined whether the relational coding of reward-associated features is commonly utilized across different learning contexts, particularly those engaging different attention modes (singleton search vs. feature-specific search) and varying levels of stimulus similarity (low vs. high target-distractor similarity). Focusing on value-driven attention based on feature relationships, our results showed that singleton search training led to value-driven relational attention that was independent of target-distractor similarity (Experiment 1a and 1b, n = 40 each). In contrast, feature-specific search training produced value-driven relational attention only when the target was dissimilar to the distractors, but not when they were similar (Experiment 2a and 2b, n = 40 each). These findings suggest a key role of the precision of target selection during reward learning in shaping value-driven attentional mechanisms. When the learning task required only coarse selection (e.g., singleton search or feature-specific search among dissimilar items), a relational code for reward-associated feature was formed; however, when fine selection was necessary (e.g., feature-specific search among similar items), a more precise code was utilized.