Abstract
Many visually-guided behaviors rely on the ability to maintain visual information in working memory. To date, however, there are few formal models of visual working memory (VWM) that directly interface with the empirical literature on this basic cognitive system. In particular, no current theories address both the maintenance of multiple items in VWM and the process of change detection within a neurally-plausible framework. Here we describe such an approach, along with data from a change detection study that confirms a novel prediction of our model.
Our model builds upon the Dynamic Field Theory of spatial working memory developed by Spencer and colleagues (Spencer & Schöner, 2003), and consists of a 1D feature WM field (e.g., color WM) coupled to a 1D “novelty” field. Neurons within each field interact via a local excitation/lateral inhibition function that allows them to form multiple, self-sustaining activation peaks in response to input. Critically, these peaks remain even when input is removed. Change detection arises from the interaction between the WM and novelty fields. These fields are coupled together in an inhibitory fashion that leads to emergent decisions about change when test stimuli are presented.
Simulations of the model have generated several novel predictions. For example, the model predicts enhanced change detection when items are highly similar— a counterintuitive prediction that has been confirmed in a recent study comparing color change detection accuracy for similar vs. distinct colors. We present these findings and contrast the DNFT with theories of VWM that rely on synchronous oscillations.