High and low: the resolution of representations in visual working memory

Abstract

Visual working memory (VWM) has long been considered to be limited in capacity, but the way in which it is limited remains unclear. One of the theoretical debates in visual working memory concerns whether the number of objects that can be stored is fixed (discrete slot models) or variable (flexible resource models). Recent research on the resolution of VWM has helped elucidate this debate by acknowledging an important trade-off between number and resolution: as the number of items stored increases, the resolution of representation declines. Yet, a different conception suggests that the number and resolution may represent distinct aspects of visual working memory, evidenced by both behavioral and neuroimaging data. In this thesis, I examined three theoretical questions regarding the relationship between the number and the resolution of items in VWM. First, how does set size affect high- & low-resolution representations (differentially)? If an item limit can be evidenced in the high-resolution measure, but not in the low-resolution measure, my second research question emerges. That is, how much resolution do we have for the remaining objects when the item limit is exceeded? Third, if both high- & low-resolution representations of an item exist in VWM, are they stored together or independently? In a series of five experiments, I addressed these questions using an adapted continuous report paradigm, in which participants were asked to remember a mixture of objects from two categories and respond firstly to the category of the item-to-report (low-resolution measure), followed by a second within-category response (high-resolution measure) which was contingent on the first. In Experiments 1-2, only performance in the low-resolution, but not in the high-resolution, measure was largely indifferent to set size, which was not compatible with either discrete slot or flexible resource models, but was largely consistent with predictions from the two-factor model and the neural object-file theory. In Experiments 3-4, precision of high-resolution representations declined monotonically until the set size reached around four items, fitting to the predictions from discrete slot models. The overall accuracy in low-resolution measure, however, remained relatively high, suggesting differential set size influence on high- and low-resolution representations. In Experiment 5, capacity comparison revealed no significant difference when the low-resolution task was absent. Taken together, I demonstrate that 1) both low-resolution ensemble representations and high-resolution individual item representations exist in VWM, and 2) high-resolution representations (i.e. object identity) and low-resolution representations (i.e. objects’ categories, configural information and perhaps some coarse feature information) of an object might be stored independently.