Accounting for stimulus-specific variation in precision reveals a discrete capacity limit in visual working memory.

Abstract

If we view a visual scene that contains many objects, then momentarily close our eyes, some details persist while others seem to fade. Discrete models of visual working memory (VWM) assume that only a few items can be actively maintained in memory, beyond which pure guessing will emerge. Alternatively, continuous resource models assume that all items in a visual scene can be stored with some precision. Distinguishing between these competing models is challenging, however, as resource models that allow for stochastically variable precision (across items and trials) can produce error distributions that resemble random guessing behavior. Here, we evaluated the hypothesis that a major source of variability in VWM performance arises from systematic variation in precision across the stimuli themselves; such stimulus-specific variability can be incorporated into both discrete-capacity and variable-precision resource models. Participants viewed multiple oriented gratings, and then reported the orientation of a cued grating from memory. When modeling the overall distribution of VWM errors, we found that the variable-precision resource model outperformed the discrete model. However, VWM errors revealed a pronounced “oblique effect”, with larger errors for oblique than cardinal orientations. After this source of variability was incorporated into both models, we found that the discrete model provided a better account of VWM errors. Our results demonstrate that variable precision across the stimulus space can lead to an unwarranted advantage for resource models that assume stochastically variable precision. When these deterministic sources are adequately modeled, human working memory performance reveals evidence of a discrete capacity limit.

Interactions between visual working memory representations

“…For example, one influential model of visual WM storage assumes that the representational quality varies randomly from trial to trial as a result of attentional fluctuations (van den Berg et al, 2012), which provides an explanation for non-normalities in the shape of the error distribution. However, as pointed out in previous studies (Bae et al, 2014; 2015; Pratte et al, 2017), different stimulus values may be represented with different degrees of precision, and trial-by-trial variability in the specific feature values provides an alternative explanation of the non-normalities. In fact, when this stimulus-specific variability is taken into account, the data are better explained by a model that does not include trial-to-trial attentional fluctuations (Pratte et al, 2017).…”

“…However, as pointed out in previous studies (Bae et al, 2014; 2015; Pratte et al, 2017), different stimulus values may be represented with different degrees of precision, and trial-by-trial variability in the specific feature values provides an alternative explanation of the non-normalities. In fact, when this stimulus-specific variability is taken into account, the data are better explained by a model that does not include trial-to-trial attentional fluctuations (Pratte et al, 2017). Here, the present study provides one more type of stimulus-specific variability— interitem interactions—that can impact the data and must be accounted for by models that attempt to explain the distribution of response errors (see also Brady & Alvarez, 2015).…”

“…The relational representation model assumes that individual items are represented in relation to each other, with each item serving as a reference for the other items. Just as perceptual representations are often biased away from reference points (Huttenlocher et al, 1991; Gibson & Rander, 1937; Pratte, Park, Rosanne, & Tong, 2017; Wei & Stocker, 2015), this model proposes that WM representations can be biased away from each other. Moreover, an attended item will be given greater weight as a reference point, so that it is not strongly biased but can more strongly bias lower-priority representations.…”

“…The present study took advantage of the fact that the perception and WM storage of orientation is strongly impacted by categorical boundaries (Huttenlocher, Hedges, & Duncan, 1991; Gibson & Rander, 1937; Girshick, Landy, & Simoncelli, 2011; Pratte, Park, Rosanne, & Tong, 2017; Wei & Stocker, 2015). Specifically, when individual orientations are perceived and remembered, their representations are repelled away from the cardinal axes (the 3 o’clock, 6 o’clock, 9 o’clock, and 12 o’clock positions on a clock face).…”

“…This model posits that each orientation serves as a reference for representing the other item. Reference points typically lead to repulsion of nearby features (Dick & Hochstein, 1989; Fisher, 1968; Gibson & Radner, 1937; Howe & Purves, 2005; Pratte et al, 2017; Wei & Stocker, 2015), so the relational representational model predicts that the reports of the two orientations will be repelled from each other, but only when they are similar (e.g., < 90° apart). This is the opposite of the prediction made by the ensemble representation models.…”