It was previously shown that the features of individual items retrieved from visual working memory (VWM) are systematically biased towards the mean feature of a sample set (Brady & Alvarez, 2011), suggesting hierarchical encoding in VWM. In our work, we investigated how hierarchical representations are stored over time. Observers were shown four differently oriented triangles for 200 ms and, after 1-, 4-, or 7-second delay, they had to report either one individual orientation, or the average orientation of all triangles, rotating a probe circle. Before set presentations, observers were informed that they had to remember one particular orientation, all four individual orientations, or the average orientation. Using the mixture model (Zhang & Luck, 2008), we estimated a probability of a tested representation being in VWM and its precision, as well as a systematic bias that would indicate hierarchical encoding. We found a strong bias towards the mean in the “remember four” condition, which provides evidence for hierarchical encoding in VWM. Our main result was the absence of significant changes in retaining the elements of a hierarchical representation (the mean and individual features). This supports an idea that hierarchical representations are related to encoding, rather than storing in VWM. Both fidelity and the probability of an item being in memory decrease over time. It supports "Sudden Death" and "Gradual Decay" accounts for storing hierarchical representations.
Previous research has documented the limited capacity of visual working memory (VWM) for color objects set at 3–5 items. Another line of research has shown that multiple objects can be stored in a compressed form of ensemble. However, existing data is more likely to testify that VWM can store no more than two such compressed units. But the nature of this discrepancy can be methodological: VWM for ensembles was never tested using methods that are applied in the research of VWM for objects. Here we have tested the capacity and precision of VWM for objects and ensembles using two standard methods — change detection and continuous report with a mixture model. We found that VWM for both types of units showed the similar capacity and precision when critical psychophysical parameters, such as foveal density and area are controlled. We also showed that this quantitative similarity between objects and ensembles is provided by a mechanism that represents each ensemble as a holistic VWM chunk as efficiently as it represents any single object.
Our interactions with the visual world are guided by attention and visual working memory. Things that we look for and those we ignore are stored as templates that reflect our goals and the tasks at hand. The nature of such templates has been widely debated. A recent proposal is that these templates can be thought of as probabilistic representations of task-relevant features. Crucially, such probabilistic templates should accurately reflect feature probabilities in the environment. Here we ask whether observers can quickly form a correct internal model of a complex (bimodal) distribution of distractor features. We assessed observers’ representations by measuring the slowing of visual search when target features unexpectedly match a distractor template. Distractor stimuli were heterogeneous, randomly drawn on each trial from a bimodal probability distribution. Using two targets on each trial, we tested whether observers encode the full distribution, only one peak of it, or the average of the two peaks. Search was slower when the two targets corresponded to the two modes of a previous distractor distribution than when one target was at one of the modes and another between them or outside the distribution range. Furthermore, targets on the modes were reported later than targets between the modes that, in turn, were reported later than targets outside this range. This shows that observers use a correct internal model, representing both distribution modes using templates based on the full probability distribution rather than just one peak or simple summary statistics. The findings further confirm that performance in odd-one out search with repeated distractors cannot be described by a simple decision rule. Our findings indicate that probabilistic visual working memory templates guiding attention, dynamically adapt to task requirements, accurately reflecting the probabilistic nature of the input.
This article reviews the research in visual working memory (VWM) over the past 20 years. We describe research methodologies in the field and focus on commonly used paradigms such as change detection and continuous report (including the use of mixed models for analysis) that aim to measure the capacity and precision of VWM. We also consider the organization of units of storage in VWM; in particular, we describe feature binding and representing multiple objects as ensemble summary statistics. We review theories that try to explain the nature of VWM limitations: structural theories (slot-based), resource theories, hybrid theories (slot and resource theories), and a recently suggested hierarchical encoding theory. Theories aiming to explain forgetting mechanisms in VWM are reviewed. We also discuss the neural correlates of VWM encoding and storage, as well as neurophysiological models of VWM that are substantially influenced by the mentioned theories.