Abstract
Our ability to extract accurate statistical mean from a set of items delivered sequentially over time is known to be limited by the degree of inter-item variability. However, neural mechanisms underlying such parametric relationship between the error of perceptual mean and inter-item variability remain poorly understood. Here we record electroencephalography (EEG) during multi-item perceptual mean orientation judgment task, and use a forward encoding model to directly recover information about the item orientation and the mean orientation of a sequence. Thus, we investigate the full representational dynamics across every single stage of perceptual mean computation of sequentially presented items as a function of inter-item variability. Observers viewed a sequence of ten randomly oriented Gabor patches presented centrally every 600 ms, and reported the orientation of their mean by adjusting a red probe bar, preceded by a blank period. Each sequence had either small or large orientation variance. During the blank period just before the perceptual mean judgment, cross-generalization between item and mean coding reveals that mean orientation is more accurately represented when the sequence is less variable and the accuracy of this recovered mean orientation also predicts behavioral performance of perceptual mean computation on a trial-by-trial basis. We also found that neural representation of mean orientation gradually becomes more precise toward the end of a sequence with small fluctuation whereas orientation encoding of each serially delivered item was equally precise regardless of sequence fluctuation. Our results indicate that the inverse relationship between the accuracy of perceptual mean and inter-item variability originates from the stage of accumulating evidence over time rather than the stage of encoding each individual item.
Meeting abstract presented at VSS 2018