Abstract
Previous studies have found that individuals maintain brain activity patterns across a working memory (WM) delay that are similar to the patterns exhibited during the initial perception of remembered items. Additionally, greater pattern similarity between encoding and recall is associated with greater chances of successful memory performance. In this fMRI study, we used a Delayed-Match-to-Sample visual WM task to investigate how ongoing changes in brain activity patterns throughout the delay interval corresponded with WM performance. On each trial, participants (n=20) viewed a target Gabor patch and were instructed to remember it using a visualization strategy throughout an eleven-second delay interval. They then saw a probe patch that either matched or did not match the target's orientation. Target and probe orientations were drawn from a set of six evenly spaced orientations, with the spacing determined by an earlier staircasing procedure to achieve approximately 75% accuracy for each participant. Non-matching probes were always chosen from an orientation adjacent to the target's. We calculated fMRI pattern similarity in visual cortex between the target representation and each subsequent timepoint in the trial. Pattern drift was defined as changes in pattern similarity during the delay interval towards or away from a given orientation's prototypical activity pattern. In trials where the target and probe orientations were the same, participants were more likely to incorrectly report the orientations did not match when their activity patterns drifted away from the target orientation and towards target-adjacent orientations. In trials where the target and probe orientations were different, participants were more likely to incorrectly report the orientations matched when their activity patterns drifted towards the orientation of the non-matching probe patch. Our results suggest that errors in working memory tasks are not simply due to unstructured noise, but also drift within representation space that can be indexed by neuroimaging.
Meeting abstract presented at VSS 2018