Abstract
Change detection tasks typically estimate adults' visual working memory (VWM) capacity to be 3–4 simple objects. To explore how capacity limits arise within a neural system, we used a dynamic neural field model of VWM to capture performance in change detection. In this model, objects are represented as “peaks” of activation that are maintained in cortical fields. Quantitative simulations show that the model must hold 5–6 peaks in VWM simultaneously to produce adult-like change detection performance. This suggests that standard methods of estimating capacity underestimate the resolution of the neural system. Moreover, the model provides novel insights into the multiple ways in which errors arise. In contrast to existing theories (e.g., Pashler's formula) which posit that false alarms reflect guesses, our model shows that false alarms reflect failures to consolidate all items from the sample array into working memory. Misses—which are not explicitly factored into capacity equations—arise due to decision errors when weak change signals are overcome by robust neural signals indicating sameness. In the final section of the presentation, we demonstrate that our model both captures existing data and generates novel predictions. In particular, the model predicts enhanced change detection for similar feature values which is consistent with recent empirical results probing change detection for both color and orientation. The model also predicts that working memories for close features will repel one another over short delays. This prediction was also successfully tested using a novel feature estimation task. Given that the model captures adults' performance and generates novel predictions, we contend it provides robust neural constraints on “capacity”, effectively grounding this concept in real-time and task-dependent neural dynamics.
Supported by NIMH RO1 MH62480 and NSF HSD 0527698 awarded to JPS.