Abstract
Working memory is our ability to temporarily hold information as needed for complex cognitive operations. Models of working memory distinguish two separate processes: (i) a selection rule that identifies the content to be recalled and (ii) the maintenance of the content. We aimed to characterize the spatiotemporal neural dynamics underlying these two components. Healthy participants performed a visual working memory task during magnetoencephalography (MEG) recording. Multivariate Pattern Analysis (MVPA) and source analyses identified two distinct types of working memory neural processes underlying selection and maintenance of the content. The selection rule is specifically decoded from sustained low-frequency (<20Hz) neural activity within a cortical network that includes the ventrolateral prefrontal cortex. By contrast, working memory content is transiently reactivated over a distributed and occipito-temporal network that differs from that encoding the sensory stimulus. These results reveal different neural mechanisms that select and maintain information in memory and could account for previous paradoxical reports of persistent and dynamic neural correlates of working memory.