Abstract
Visual short-term memory (VSTM) performance as a function of set size is well accounted for by noise corrupting the stimulus representation, with the amount of noise increasing with set size. It has been proposed that, in addition to this mechanism, there is also a loss of binding between feature and location information (Bays et al, 2009). An analysis of delayed-estimation data suggests that the prevalence of such binding errors is low (Van den Berg, Awh, and Ma, 2014), but this analysis was quite indirect. Here, we address the question of whether location information is maintained in VSTM with a more direct approach. 11 observers performed two VSTM-based tasks with arrays of 2,3,4 and 6 items: a target detection task (target present half of the time) and a target localization task (always one target). Any loss of location information would affect localization performance but not detection performance. Therefore, if we can jointly fit an optimal observer model with the same parameters to detection and localization, this would suggest that location information loss is minimal. Indeed, we were able to fit well the variable-precision encoding model jointly to the detection and localization data. These preliminary model fits suggest that location information is maintained in VSTM to a significant extent.
Meeting abstract presented at VSS 2017