Abstract
Clustering is an efficient strategy to represent redundant information with limited memory resources. If apples have similar shapes and colors in an orchard scene, the visual system may encode a cluster of apples rather than each different apple's individual shape and color. The current study investigated the impact of clustering on representational quality of visual working memory (VWM). We hypothesized that similar items are organized into a cluster, and their recall precision becomes higher with fewer clusters. Across four experiments, we manipulated the orientation similarity of several bars so that they formed a different number of clusters. Participants remembered bar orientations and later estimated the orientation(s) of cued bars. Based on these estimations, we measured recall bias to confirm that similarly oriented bars formed a cluster, and measured recall precision to test whether the recall quality improved with fewer clusters. Specifically, in Experiments 1 and 2, five bars formed one, two, or three clusters. In Experiment 3, five bars formed two clusters, or three bars each formed their own clusters. Consistent with the prediction, similar orientations were recalled with a bias toward their mean orientation, indicating that similar items formed a cluster. Also, recall precision of each item was higher with fewer clusters, regardless of the number of individual items. In Experiment 4, we parametrically manipulated the similarity between cluster means and measured response correlation in addition to recall bias and precision. We found that response correlation was not influenced by the similarity between the cluster means. This result indicates that it is the number of clusters itself that has a critical role for VWM quality, rather than the similarity between clusters. Taken together, we suggest that clusters formed by similar items impact VWM representation and its quality, acting as representational units of VWM.
Meeting abstract presented at VSS 2018