The neural signature of information regularity in temporally extended event sequences

Statistical regularities exist at different timescales in temporally unfolding event sequences. Recent studies have identified brain regions that are sensitive to the levels of regularity in sensory inputs, enabling the brain to construct a representation of environmental structure and adaptively generate actions or predictions. However, the temporal specificity of the statistical regularity to which the brain responds remains largely unknown. This uncertainty applies to the regularities of sensory inputs as well as instrumental actions. Here, we used fMRI to investigate the neural correlates of regularity in sequences of task events and action selections in a visuomotor choice task. We quantified timescale-dependent regularity measures by calculating Shannon's entropy and surprise from a sliding-window of consecutive task events and actions. Activity in the frontopolar cortex negatively correlated with the entropy in action selection, while activity in the temporoparietal junction, the striatum, and the cerebellum negatively correlated with the entropy in stimulus events at longer timescales. In contrast, activity in the supplementary motor area, the superior frontal gyrus, and the superior parietal lobule was positively correlated with the surprise of each stimulus across different timescales. The results suggest a spatial distribution of regions sensitive to various information regularities according to a temporal hierarchy, which may play a central role in concurrently monitoring the regularity in previous and current events over different timescales to optimize behavioral control in a dynamic environment.