Abstract
Humans allocate visual working memory (VWM) resource according to behavioral relevance, resulting in more precise memories for more important items (Bays & Husain, 2008; Emrich et al., 2017; Klyszejko et al., 2014; Yoo et al., 2018). Theoretically, items may be maintained by feature-tuned neural populations, where the relative gain of the populations encoding each item represents precision (Bays, 2014; Ma et al, 2006). To test this hypothesis, we compared the amplitudes of neural activity in the different parts of retinotopic maps representing each of several VWM items. We predicted that the magnitude of delay period activity in areas topographically associated with different items would monotonically track the priority of those items. We scanned participants with fMRI while they performed a visuospatial WM task. Participants remembered the location of four items, one presented in each visual field quadrant, then generated a memory-guided saccade to a probed item after a 10 second delay. Before the four items appeared on each trial, a cue indicated the probability with which each item would be probed for response (0.6, 0.3, 0.1, 0.0). We measured fMRI activity in topographically organized dorsal stream areas known to be important for VWM. In a separate session, we defined visual field maps in occipital, parietal, and frontal cortex using a modified population receptive field mapping technique (Mackey, Winawer, & Curtis, 2017). Behaviorally, we found that the precision of VWM scaled monotonically with the priority of the item, replicating our previous work. Neurally, the amplitude of BOLD activation within voxels corresponding to the retinotopic location of VWM items scaled monotonically with the priority of the item. These results suggest that the distribution of WM resource according to priority sculpts the relative gains of neural populations that encode items with varying precision.