Abstract
Two distinct kinds of information can be represented in visual short-term memory: precise representations of individual objects (individuation), and summaries/averages of multiple objects (ensemble representation). Prior studies have found that observers can accurately report average features (e.g., location or facial emotion) from sets of many items even when unable to recall the features of any individual item, suggesting that individuation and ensemble representation may be subserved by distinct processes, and perhaps also distinct memory resources. Here, we suggest that these modes of representation may instead rely upon a shared pool of memory resources, by demonstrating that the precision of individuation and ensemble representation are mutually interdependent: precise encoding of one type of information entails reduced precision in the other. In Study 1, observers briefly viewed three discs and were subsequently cued to report either a specific disc's location (Individuation task) or the centroid of all three discs (Ensemble task). On some trials, the discs were also connected by bars. We reasoned that connection should bias the visual system towards storing the centroid of the larger object that it creates, whereas in unconnected displays the bias should be towards representation of discrete individuals. As predicted, connection produced enhanced memory for the centroid and impaired memory for individual object locations (supported by a highly significant Connectedness-by-Task interaction). In Study 2, we tested whether observers could switch between individuation and ensemble representation when given pre-cues that predicted (with 80% validity) the subsequent test type. Valid pre-cues led to enhanced accuracy on both tasks, suggesting that observers can effectively control the preferred encoding. Collectively, these findings suggest that individuation and ensemble representation compete with one another, and that the visual system can flexibly switch between these modes as a way of efficiently managing its limited memory resources.
Meeting abstract presented at VSS 2014