Abstract
Working memory (WM) is a core cognitive function that enables the maintenance of information no longer present in the environment for guidance of behavior. When more information must be maintained, recall precision is impaired (e.g., Bays & Husain, 2008). Using a multivariate image reconstruction technique (Sprague & Serences, 2013), we recently demonstrated that item representations carried by fMRI activation patterns degrade with increasing memory load (Sprague, Ester & Serences, 2014). However, it is not clear if degraded item representations are permanently corrupted due to rapid degradation at encoding, or whether multiple representations mutually suppress each other during retention. If the latter were true, removing an item from WM midway through a long delay interval would resuscitate the fidelity of a degraded representation. To evaluate this possibility, we required participants to precisely maintain either one or two spatial positions in WM during a long delay interval (16 s). On half of trials when two items were maintained, participants were post-cued 8 s into the delay about which item would be queried at the end of the trial with 100% validity, so that they could “drop” the irrelevant item from WM. Though WM recall precision was behaviorally impaired with increasing set size, when a second item was dropped recall performance recovered to that when a single item was maintained. We computed spatial reconstructions using activation patterns from both the first and second half of the WM delay interval. After one item had been dropped from WM, we observed robust recovery of the remaining target representation over the dropped target representation. Together, these results suggest that multiple simultaneous representations can mutually inhibit each others’ fidelity (Franconeri et al, 2013). When one of several items is removed from WM, the normalization drive is reduced, allowing the remaining item representation to rebound in fidelity.
Meeting abstract presented at VSS 2015