Working memory improves developmentally as neural processes stabilize.

Working memory performance is a key indicator of cognitive and developmental status. While recent evidence indicates that stabilizing neural gain supports the stabilization of working memory during adolescence, the computational mechanisms linking neural stabilization to behavior are poorly understood. We develop a mechanistic account of behavior during the memory-guided saccade task based on a stochastic accumulator framework. Results indicate that a specific balance of independent gain signals affecting working memory representations and oculomotor response thresholds can account for a peculiar U-shaped feature of the speed-accuracy relationship. Additionally, aspects of behavioral variability and mean behavioral performance, as well as subtle shifts in the shape of the speed-accuracy relationship across development, can be accounted for by the stabilization of these two sources of variability. Thus, the stabilization of neural variability can, in part, account for developmental improvements in behavioral variability as well as some improvement in mean behavioral performance.