Abstract
The contents of visual working memory (WM) can be decoded from the spatial patterns of delay period fMRI activation in occipital, parietal, and frontal cortices (Serences et al, 2009; Harrison & Tong, 2009; Jerde et al, 2012; Lee et al, 2013; Albers et al, 2013; Sprague et al, 2014; Ester et al, 2015; van Bergen et al, 2015; Rahmati et al, 2017). Here, using fast-sampling (750 ms) fMRI, we characterized the temporal dynamics of these WM representations across the human cortex from initial encoding to the end of maintenance. On each trial, participants generated a memory-guided saccade to the location of a target stimulus briefly presented prior to a 12-second-long retention interval. We applied a linear inverted encoding model (IEM) to reconstruct the remembered spatial position at each time point during the trial from cortical activation patterns (Sprague et al, 2014), focusing analyses on retinotopic regions independently-defined using voxel receptive field mapping in occipital, parietal, and frontal cortex (Mackey et al, 2017). We observed an initially-strong representation in V3AB, followed by simultaneous cascades backward from V3 to V2 to V1 and forward along retinotopic IPS over the next several seconds. In many regions, representations remained stable throughout the delay period, but waxed and/or waned in their strength. We confirmed this stability by evaluating the extent to which an IEM estimated using data from each time point could generalize to others (King & Dehaene, 2014). We found evidence for stable representational geometry across the entire 12 s delay in extrastriate visual and parietal cortex, further supporting the notion that WM representations are instantiated via persistent stable codes. This dynamic routing of information suggests a nuanced perspective on the role of different brain regions during WM maintenance, demonstrating that each region represents information most strongly at different points in time.
Meeting abstract presented at VSS 2018