Visual Working Memory: Space, features

Talk Session: Sunday, May 17, 2026, 10:45 am – 12:30 pm, Talk Room 2
Moderator: Freek van Ede, Vrije Universiteit Amsterdam

Schedule of Events | Search Abstracts | Symposia | Talk Sessions | Poster Sessions

Talk 1, 10:45 am, 32.21

Suppression of Passive Memory Reflects Proactive and Context-Driven Control in Visual Working Memory

Ziyuan Li¹ (), Mengyao Sun¹; ¹Anyang Normal University

Hierarchical models of visual working memory (VWM) propose that information can be retained in multiple functional states, with currently relevant representations maintained in an active state and prospectively needed representations stored in a passive state. Recent work shows that passive memories undergo sustained suppression, a mechanism that stabilizes passive maintenance and prevents interference with concurrent active processing. However, it remains unclear whether such suppression is obligatorily triggered by the actual demands of a secondary task or reflects an expectation-driven cognitive strategy. To address this question, we designed two tasks performed in a counterbalanced order while recording EEG: (1) a baseline task where memory items (array 1) were always relevant and maintained in the active state until retrieval; and (2) a probability task where array 1 was followed either by a blank retention interval (30%) or by frequent presentation of a second array (array 2; 70%). In this probability task, array 1 became prospectively relevant and was expected to transition to a passive state when array 2 required active maintenance. When participants firstly habituated to active maintenance in the baseline block, suppression of array 1 emerged only when array 2 was actually presented, indicating that suppression of passive maintenance was driven by actual processing demands. In contrast, when the baseline mindset was removed, participants adopted a proactive strategy of switching array 1 into a passive state regardless of whether array 2 appeared; this suppression was flexibly withdrawn once participants has no anticipation of additional processing. Behaviorally, switching memory state incurred performance cost. Together, these findings demonstrate that the suppression of passive memory is not purely stimulus-driven, but reflects an expectation-guided mindset preparatory for anticipated processing demands. This study reveals that state transitions in VWM are dynamically governed by learned task structure, underscoring the proactive and context-sensitive nature of human memory control.

National Natural Science Foundation of China

Talk 2, 11:00 am, 32.22

The Causal Roles of Visual and Prefrontal Cortex in Visual Working Memory Prioritization: A Phosphene- and MRI-guided rTMS-EEG Study

Chenlingxi Xu¹ (), Nathan Rose²; ¹University of Notre Dame, ²Northeastern University

While the frontal cortex is traditionally viewed as the hub for working memory (WM) maintenance, recent theories predict that sensory cortex [early visual cortex (VC)] assists in the maintenance of unattended memory items via “activity-silent” short-term synaptic plasticity. Prior studies have implicated interregion communication between both dorsolateral prefrontal cortex (dlPFC) and VC during WM maintenance. Here we propose a functional division of labor: dlPFC controls attentional priority and interference resolution whereas VC maintains high-fidelity sensory traces. Causal evidence dissociating these contributions during dynamic prioritization is limited. Sixty-eight healthy younger adults performed a two-item, double-retrocue, double-recall orientation visual WM task with concurrent EEG recording before and after receiving offline active or sham rTMS (40-s) at either theta-burst (cTBS) or 10 Hz (alpha) stimulation to MRI-guided left dlPFC or phosphene-localized left VC. Behavioral performance was quantified using absolute recall errors and computational modeling (precision, swap errors, guessing). Our pre-registered predictions are that, relative to sham, 10 Hz-rTMS to dlPFC will increase swap error rates, while cTBS to VC will decrease the precision of orientation recall. Preliminary analyses of the active stimulation group reveal emerging site-specific patterns. Following dlPFC stimulation, participants showed general performance gains across trial types. In contrast, VC stimulation showed hemifield-dependent effects: improvements were attenuated when the retrocued item appeared in the TMS-targeted hemifield. Furthermore, a site-by-frequency interaction was observed for precision during attention switching: 10-Hz-stimulation decreased precision when applied to dlPFC but increased precision when applied to VC. While active stimulation drove overall gains rather than predicted impairment, the dissociation between general (dlPFC) and hemifield-specific (VC) patterns were consistent with the hypothesized division of labor. Planned EEG analyses will further evaluate whether dlPFC stimulation alters control-related signals (e.g., N2pc, P3, midfrontal theta) and VC stimulation alters sensory-maintenance signals (e.g., N1/P1, CDA, posterior alpha), providing potentially converging mechanistic evidence.

Talk 3, 11:15 am, 32.23

Contralateral delay activity and alpha lateralisation during working memory are contingent on the transition from sensation to working memory

Baiwei Liu¹, Freek van Ede¹; ¹Vrije Universiteit Amsterdam

Delay-period EEG signals, such as the Contralateral Delay Activity (CDA) or the lateralisation of 8-12 Hz posterior alpha activity, are widely viewed and used as direct markers of selective working-memory maintenance. However, these EEG markers are conventionally studied in pre-cue paradigms where relevant memory content must first be selected at sensory encoding. This potentially conflates the selective maintenance of relevant information with the active translation of this information from sensation into memory. To disentangle selective sensory-to-memory transition from selective memory maintenance, we leveraged a novel task where we could compare situations that were otherwise identical, but where in one case relevant memory content had to first be selected from perception, while in the other case this content could be selected from working-memory directly. In both cases, the selected visual content and the ensuing working-memory task were identical, but only the former engaged a sensory-to-memory transition. Strikingly, our results show that sustained delay activity was contingent on the availability of sensory input: both the CDA and prolonged alpha lateralisation were robust in the ‘select at encoding’ condition but absent in the ‘select from working memory’ condition. Crucially, this was not due to a lack of spatial organisation in the latter ‘select from working-memory’ condition, as we observed robust and comparable N2pc and initial alpha lateralization in both conditions. This result was further supported with multi-variate pattern analysis. Together, our results show how prolonged “delay activity” EEG signatures in posterior electrodes are surprisingly contingent on sensory input, and suggest that such signatures may track the transition from sensory inputs to working memory traces, rather than the pure maintenance of relevant information per se.

Talk 4, 11:30 am, 32.24

Electroencephalogram decoding suggests separate indexing mechanisms for attentional tracking of featureless objects and working memory storage

Henry Jones¹ (), Dawei Bai², Brian Scholl², Edward Awh¹; ¹University of Chicago, ²Yale University

Recently, multivariate decoding of EEG data has identified a signal that scales with the number of items in working memory (WM), regardless of the specific content being maintained or the number of spatially attended locations. One possibility is that this signal reflects an abstract indexing process that binds the content of items to their context in space and time for maintenance and accessibility. To explore this possibility, we examined whether a similar load signal exists for “featureless objects”, which can only be described by their coordinates in space and time. On each trial, participants viewed a dense grid of crosses of random orientations. A moving object was implemented by having a single cross change from one random orientation to another, with changes occurring between adjacent crosses across frames. These transients yielded a persisting trackable object, even though (a) there is no constant surface feature across frames, and (b) it is impossible to identify objects on any static frame. In 2 EEG datasets, participants completed a task in which they tracked either 1 or 2 cued featureless objects, and a WM task in which they remembered either colors or the shape and location of “dot cloud” stimuli while spatial attention was controlled for. In both experiments, EEG decoding found a stable signal that scaled with the number of tracked featureless objects. However, we found no consistent evidence that the tracking load signal and the WM load signal generalized to one another. Although planned studies will examine whether this conclusion generalizes to traditional tracking tasks that may rely on other mechanisms (Lu and Sperling, 1996), these results suggest that distinct mechanisms may guide WM storage and attentional tracking.

Talk 5, 11:45 am, 32.25

Shifts of spatial attention within working memory occur faster than feature re-prioritization.

Caitlin V. Cunningham¹, Julie D. Golomb¹; ¹The Ohio State University

In our complex visual world, spatial attention is used to boost processing for objects at an attended location. In addition to operating on items in the external environment, spatial attention can be directed internally, toward items within working memory. Spatial attention is frequently reoriented to keep up with the dynamic environment, but these attentional shifts are imperfect: e.g., shifts of externally-directed spatial attention are known to result in errors in feature perception. Here, we seek to track the dynamics and consequences of shifting internal attention. In Experiment 1, participants encoded an array of memory items (colored squares), then were peripherally cued to orient internal attention toward one item’s location. On some trials, a second postcue subsequently cued participants to shift spatial attention to a different item within working memory. After a short (50ms) or long (500ms) delay, participants reported the color and location of the most recently cued item (target). Probabilistic mixture modeling revealed that when given 50ms to shift attention, participants almost always reported the correct location of the target item, but sometimes made feature errors: specifically, they tended to misreport the color of the item opposite the first-cued item in color space, characteristic of “feature avoidance errors.” These errors indicate that participants were aware they were supposed to report a different item than initially cued, but couldn’t yet access the feature, so strategically guessed a color maximally different from the color still “stuck” in memory. We replicated this pattern in Experiment 2, which was identical to Experiment 1 but with endogenous postcues presented at fixation instead of peripherally, revealing that these errors are not driven by eye movements, but indeed associated with failures to shift covert internal attention. Together, these results suggest that shifting of internal spatial attention occurs more quickly than re-prioritization of features in working memory.

NIH R01-EY025648 (JG)

Talk 6, 12:00 pm, 32.26

Visual working memory across saccades: evidence for independent pre- and post-saccadic representations

Zexuan Niu¹, Andrew Hollingworth¹; ¹UNIVERSITY OF IOWA

Before the saccade to an object, perceptual information is acquired peripherally and is maintained in visual working memory across the saccade. Post-saccadic information is then encoded, typically from the now-foveated object. How do the pre- and post-saccadic representations interact in object perception? Recent work suggests that the two representations are integrated into a single, composite representation (the Integration Hypothesis), supported by evidence that mean feature report lies at an intermediate value between pre- and post-saccadic values. A plausible alternative, however, is that each representation is maintained independently (the Independence Hypothesis), with participants probabilistically reporting either the pre- or post-saccadic value. In our first experiment, participants made a saccade to a colored disk, whose color changed transsaccadically. They then reported the pre-saccadic color. The integration hypothesis predicts a univariate distribution of color report, centered between the pre- and post-saccadic values. In contrast, the independence hypothesis predicts either 1) a bivariate distribution, with means fixed at the pre- and post-saccadic values or 2) a univariate distribution centered at either the pre- or post-saccadic value. In our data, the independence model generally provided a better fit to the observed distribution of color report than did the integration model, although individual differences emerged with larger color change. In a second experiment, participants indicated whether they perceived one or two colors, and the analyses were restricted to single-color trials to minimize the influence of explicit change detection or strategic inference. Here, the independence model consistently provided a better fit than did the integration model. Typically, participants either retained both pre- and post-saccadic information (i.e., a bivariate response distribution) or post-saccadic input overwrote pre-saccadic information (i.e., a univariate response distribution centered at the post-saccadic value). Together, the data suggest that pre- and post-saccadic values are not directly integrated across saccades and are instead maintained independently.

Talk 7, 12:15 pm, 32.27

Motoric mechanisms that bias gaze produce reflexive selection in memorized visual space

Priyanka Gupta¹, Devarajan Sridharan¹; ¹Indian Institute of Science, Bangalore

Although the shared neural mechanisms of eye-movements and covert attention are well-studied, the link between eye-movements and working memory (WM) remains tentative. Whether and how gaze mechanisms affect selection in visual WM is actively researched. Here, we present a cognitive mechanism that biases gaze and produces reflexive (automatic) selection in memorized visual space. We tested the effect of "saccadic tagging" – cueing a location for a subsequent saccade – on spatial selection in WM. In one dual-task, participants (n=24) executed cued saccades to one of four locations (Go), while also maintaining in WM orientations of gratings presented at each of those locations. On 40% of trials, participants were strongly incentivized to plan saccades but refrained from executing them (No-go). Saccadic tagging produced robust spatial selection in WM, quantified by the mean absolute error (MAE) of orientation recall (cued: 29.39°±1.51°, uncued: 32.49°±1.27°, p<0.001). We note three remarkable aspects of this selection. First, spatial selection occurred for the tagged location regardless of whether the saccade was executed (Go) or merely planned (No-go) toward it. Second, selection occurred despite each of the four locations being equally task-relevant (equal recall probe validity). Finally, nearly identical selection occurred also in a manual response task designed to avoid systematic spatial correspondences between the cue and button locations. In sum, modality-generic motoric preparation sufficed to produce robust, reflexive selection in memorized visual space. We sought putative physiological underpinnings of this mnemonic selection. Microsaccades – small, fixational eye-movements – were strongly biased toward the tagged location (p<0.001), across both modalities. Moreover, the direction of the first microsaccade post-cue predicted the extent of reflexive selection in WM, irrespective of motor modality. Our results link motoric mechanisms that bias gaze to those that produce selection in visual WM. The findings may be relevant for diagnosing and treating disorders of WM.