Purchase this article with an account.
Maija Honig, Daryl Fougnie, Wei Ji Ma; Probabilistic Information in Visual Working Memory. Journal of Vision 2016;16(12):1054. doi: 10.1167/16.12.1054.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Current theories assume that a stimulus is stored in working memory as a point estimate, e.g. a specific color value. Another possibility is that memory representations are richer, perhaps containing a probability distribution on every trial. To test for this possibility, we conducted a delayed-estimation experiment in which subjects reported both a stimulus estimate and an uncertainty value. They remembered the colors of four items, after a delay estimated the color of a randomly chosen item on a continuous color wheel, and finally adjusted the size of an arc over that wheel centered on the estimate. When the true color was "captured" by the arc, the subject received points; the points decreased linearly with increasing arc size. This incentivized subjects to report an arc size that reflected their memory uncertainty on that trial. To examine in more detail how people utilized memory uncertainty, we introduced prior beliefs about likely colors by training subjects on one of two color distributions: either uniform (all colors are equally likely) or Von Mises (some colors are more frequent than others). We found that when subjects reported a larger arc size (higher uncertainty), their error was higher, suggesting that they possessed meta-knowledge of memory quality. Moreover, when colors came from a Von Mises distribution, reports were biased towards the most frequent color and the amount of bias increased with increasing arc size. These patterns of results are consistent with a Bayesian-observer model in which memory uncertainty is represented, varies across trials, and gets combined with prior beliefs and reward information to produce a response. These results support that probabilistic information is stored in visual working memory, and that people have access to this information and incorporate it into decisions.
Meeting abstract presented at VSS 2016
This PDF is available to Subscribers Only