Purchase this article with an account.
Hsin-Hung Li, Thomas Sprague, Aspen Yoo, Wei ji Ma, Clayton Curtis; Neural correlates of uncertainty of visual spatial working memory in human cortex. Journal of Vision 2020;20(11):1320. doi: https://doi.org/10.1167/jov.20.11.1320.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Humans have knowledge about the quality of their visual working memory (VWM) representations (Rademaker et al., 2012; Yoo et al., 2018). This knowledge informs humans to what extent they should trust their memory when making decisions. Using a Bayesian generative model that considered both the neural and fMRI measurement noise (van Bergen et al., 2015), we previously decoded spatial VWM content from fMRI BOLD signals (Sprague et al, VSS 2017; 2019). For individual trials, this analysis extracted posterior distributions over feature values, which generated accurate estimates of remembered locations (posterior mean). Additionally, decoded neural uncertainty (standard deviation of the posterior) correlated with the VWM precision. While the previous work revealed the relation between neural uncertainty and memory precision, the neural correlates of VWM uncertainty explicitly reported by humans remain unknown. We investigated this issue in a spatial VWM task: A target dot (12° eccentricity) appeared at a randomly chosen location in each trial. After a delay, observers made a saccade to the remembered target location, and then performed a post-decision wagering task by adjusting the length of an arc presented around the reported location. Observers received points if the true target location fell within the arc, but the number of points decreased with arc length. This procedure incentivized observers to scale the arc length according to their uncertainty of VWM. We found that the posterior distribution decoded from fMRI data accurately estimated remembered locations. In the wager task, the arc length correlated with the magnitude of memory error, supporting its utility as a measure of uncertainty. Moreover, the arc length correlated with decoded neural uncertainty in V3AB and IPS, but not in early visual cortex (V1-V3). These results highlight the role of higher-level visual and parietal cortex in representing the contents of VWM in a format that reflects uncertainty.
This PDF is available to Subscribers Only