Figure 2. Schematic of signal amplitudes in the dynamic neural resource (DyNR) model during a cued recall trial.

(A) Observers are presented with a memory array (left), followed after a blank delay (not shown) by an arrow cue (center) indicating the location of one item (the target) whose remembered orientation should immediately be reported (right). (B) The amplitude of the visual input associated with each item is modeled as a step function (left). The sensory response (D) is modeled as a low-pass filtering of the stimulus input, with different time constants for rise and decay (C). (F) Amplitude of the working memory signal reflects a saturating accumulation of activity from the sensory population (illustrated in E). Beginning with stimulus onset, activity associated with each item is accumulated from the sensory population into the visual working memory (VWM) population, approaching an upper bound (green dashed line) that reflects a total activity limit shared between the ${\displaystyle N}$ items in memory. Once the cue has been presented (solid orange line) and processed (dashed orange line), uncued items can be dropped from VWM, raising the ceiling on activity available to represent the cued item (green arrow). This allows more information about the cued item to be accumulated from the decaying sensory trace (equivalent to the red shaded area in D). Response variability depends on the asymptotic VWM signal amplitude available for decoding (red circle) combined with the accumulated effects of diffusion (see text).