Abstract
There is general agreement that some form of sustained activation is the most plausible neuronal substrate for maintenance in working memory (WM). In the present study, we describe a dynamic neural field (DNF) model of WM that achieves a stable memory state through locally excitatory and laterally inhibitory interactions among feature-selective populations of neurons. This form of interaction allows self-sustained peaks of activation to be maintained in the absence of input (i.e., after the stimulus is removed). However, this can also give rise to metric-dependent interactions among peaks when more than one item is being held in WM. One consequence of such interactions, which we explore here, is that close peaks in WM (e.g., similar colors) will repel each other over delays. This arises as a result of shared lateral inhibition between nearby items in WM. Specifically, when two similar items are maintained, the lateral inhibition in-between them is greater than the inhibition on the “outside” of each peak, causing them to drift away from each other over the delay. To test this prediction, participants completed a color estimation task probing WM for color. On each trial, participants were shown a memory display that contained two ‘close’ colors and one ‘far’ color. After a brief delay, a color wheel was presented at one of the three target locations, cueing the participant to estimate the color that was originally at that location. The DNF model predicts that estimates of the close colors should be biased away from each other across the delay, whereas estimates of the far color should be comparable to performance when only a single color was remembered. Results confirmed this prediction, suggesting that items in WM interact in a metric-dependent fashion. We discuss the implications of these findings for other models of working memory.