Searching for targets in visual working memory: investigating a dimensional feature bundle (DFB) model

Abstract

The human visual working memory (WM) system enables us to store a limited amount of task-relevant visual information temporally in mind. One actively debated issue in cognitive neuroscience centers around the question of how this WM information is maintained. The currently dominant views advocated by prominent WM models hold that the units of memory are configured either as independent feature representations, integrated bound objects, or a combination of both. Here, we approached this issue by measuring lateralized brain electrical activity during a retro-cue paradigm, in order to track people's ability to access WM representations as a function of the dimensional relation between WM items and task settings. Both factors were revealed to selectively influence WM access: whereas cross relative to intradimensional WM targets gave rise to enhanced contralateral delay activity (CDA) amplitudes, localization relative to identification task demands yielded speeded CDA and manual response times. As these dimension-based findings are not reconcilable with contemporary feature- and/or object-based accounts, an alternative view that is based on the hierarchical feature-bundle model is proposed. We argue that WM units may consist of three hierarchically structured levels of representations, with an intermediate dimensionally organized level that mediates between top-level object and lower-level feature representations.

Electroencephalographic evidence for improved visual working memory performance during standing and exercise

“…Following the jackknife method, F‐values were corrected using the formula: F Corrected = F /( n ‐1) 2 (Miller et al ., ). It should be noted that differences observed in the onset and offset timing of averaged ERL waveforms can potentially be due to differences in trial‐to‐trial variance within the respective conditions (e.g., Töllner et al ., ). To control for this, we further calculated the CDA and LRP width (offsets‐minus‐onsets) for each condition.…”

“…The aim of the present study was to investigate the influence of moderate acute aerobic exercise and body posture on the temporal dynamics of concurrent VWM performance. To this end, participants performed a VWM retro‐cue task adopted from Töllner, Eschmann, Rusch, and Müller (), Töllner, Mink, and Müller () concurrently during conditions of rest and exercise across two postural modalities: seated on or pedalling a stationary bicycle, as well as standing or walking on a treadmill. Along with behavioural data, EEG was recorded with a focus on segregating and isolating three temporally consecutive and functionally distinct stages of the VWM processing pipeline: accessing actively held target representations (CDA), selecting a motor response (sLRP), and executing the selected motor response (rLRP) – providing a means to discern when and where exercise‐driven cognitive modulations may occur (see Figure for an illustration of the processing pipeline).…”

Attending to multiple objects relies on both feature- and dimension-based control mechanisms: Evidence from human electrophysiology

“…On the basis of these results, we recently introduced the notion of "dimensional feature bundles" (DFB) (Töllner, Mink, & Müller, 2015), which extends the well-known, hierarchically structured feature bundle model (Brady, Konkle, & Alvarez, 2011) by adding an intermediate, dimension-based level of WM representations (that connects top-level object with lower-level feature representations). Accordingly, increased processing times for cross-dimensional (relative to intra-dimensional) target processing in the retro-cue task can be explained by the requirement to actively maintain and scan two (instead of just one) dimensionally organized feature bundles (for further details, see Töllner, Mink, et al, 2015). Whereas the DFB model was originally devised to explain RT differences in memory search, in theory, the same dimension-based processing dynamics also may be at work in the current paradigm.…”

Introduction to Competitive Visual Processing Across Space and Time: Attention, Memory, and Prediction

“…To address the representational format of working‐memory contents, Töllner, Mink, and Müller combine an EEG measure with behavioral measures. They conclude that there are three hierarchical stages of memory representation, representing objects, dimensions, and features, respectively …”

“…They conclude that there are three hierarchical stages of memory representation, representing objects, dimensions, and features, respectively. 21 Foerster and Schneider examine the effect of long-term memory on selection in a task that requires learning and reproducing an action sequence. Specifically, they ask how overt attentional control by longterm memory is interrupted in the face of an unpredicted target-sequence change, and find evidence for a shift from memory-based selection to visual search.…”