Abstract
We have a good understanding of how well people can remember a relevant feature of an attended object (e.g. color of a color patch; Bays et al., 2009), but we know relatively little about how an attended object’s task irrelevant features are encoded (e.g. shape of a color patch). Previous research suggests that we have some categorical knowledge of irrelevant features (Eitam et al., 2013), but knowing the precision with which this information is stored is essential for understanding the capacity of working memory. To better understand the memory of irrelevant features, we presented participants with a colored arrow, and for the first 25 trials, consistently asked participants about its color using a color wheel. On trial 26, we cued participants to recall the direction of the arrow in a surprise test (i.e. the critical trial). To determine how participants are responding on the critical trial, we compared the log likelihood of different combinations of von Mises distributions’ fit to the behavioral data, with the distributions varying from precise to guessing (i.e. uniform distribution). From this comparison process, we find that some participants remember the irrelevant information imprecisely, relative to a relevant feature response distribution, while other participants are simply guessing. Importantly, after the surprise test, all participants could easily recall the arrow’s direction but this improvement in direction memory came at a significant cost in precision for color memory. We attribute these findings to varying levels of attention to different features during memory encoding. To understand how these tradeoffs occur at a mechanistic level, we simulated featural attention in the Binding Pool model of visual Working Memory (Swan & Wyble 2014) according to findings from monkey neurophysiology (McAdams et al., 1999). The model accurately simulates imprecise retrieval for irrelevant features and the cost of adding features to a memory.
Meeting abstract presented at VSS 2015