Discrete fixed-resolution representations in visual working memory

Abstract

Limits on the storage capacity of working memory significantly affect cognitive abilities in a wide range of domains, but the nature of these capacity limits has been elusive. Some researchers have proposed that working memory stores a limited set of discrete, fixed-resolution representations, whereas others have proposed that working memory consists of a pool of resources that can be allocated flexibly to provide either a small number of high-resolution representations or a large number of low-resolution representations. Here we resolve this controversy by providing independent measures of capacity and resolution. We show that, when presented with more than a few simple objects, human observers store a high-resolution representation of a subset of the objects and retain no information about the others. Memory resolution varied over a narrow range that cannot be explained in terms of a general resource pool but can be well explained by a small set of discrete, fixed-resolution representations.

Fig 1

a, Colour recall task. b, Mixture model of performance, showing the probability of reporting each colour value given a sample colour at 180°. When the probed item is present in memory, the reported colour tends to be near the original colour (blue broken line). When the probed item is not present in memory, the observer is equally likely to report any colour value (red broken line). When collapsed across trials, the data comprise a mixture of these two trial types (solid line), weighted by the probability that the probed item was stored in memory. c, Results of Experiment 1 (N=8).

Fig 2

P_m and SD results from Experiment 2 (N=8). The lines show the SD predictions of a pure resource model (black dashed line), the slots+averaging model (gray solid line), and the slots+resources model (black dotted line). Error bars show within-subjects 95% confidence intervals.

Fig 3

a, Stimuli and results from Experiment 3 (N=22). The cue appeared simultaneously with the sample array. b, Stimuli and results from Experiment 4 (N=8). A colour wheel and probe array appeared at the end of the trial, 900 ms after sample offset. Error bars show within-subjects 95% confidence intervals.