Global Effects of Feature-based Attention Depend on Surprise

Abstract Recent studies have shown that prediction and attention can interact under various circumstances, suggesting that the two processes are based on interdependent neural mechanisms. Attention is broadly used to describe a number of subcomponents, however, and it remains unclear which of these mechanisms interacts with prediction and in what manner. In particular, it is currently unknown whether prediction can interact with feature-based selection outside the current spatial focus of attention. Here, across two experiments, we recorded neural activity with electroencephalography (EEG) as human observers performed a feature-based attention task at fixation and ignored a stream of peripheral stimuli with predictable or surprising features. Central targets were defined by a single feature (colour or orientation) and differed in salience across the two experiments. Task-irrelevant peripheral patterns usually comprised one particular conjunction of features (standards), but occasionally deviated in one or both features (deviants). Consistent with previous studies, we found main effects of feature-based attention and prediction on neural responses to task-irrelevant patterns in both experiments. Crucially, we observed an interaction between the effects of prediction and feature-based attention in both experiments: neural responses to surprising but not predictable patterns were modulated by feature-based attention. These findings suggest that global effects of feature-based attention depend on surprise, and are consistent with the idea that attention optimises the precision of predictions by modulating the gain of prediction errors.