Pinging the brain to reveal the hidden attentional priority map using encephalography

The study addresses how past experiences (selection history) shape current attentional deployment through a latent attentional priority map. The authors outline how the brain leverages statistical learning to predict relevant information while filtering out irrelevance, integrating with selective attention to optimize performance. Although priority maps have been extensively linked to top-down and bottom-up influences, selection history’s neural basis appears activity-silent, likely mediated by synaptic-level changes rather than sustained neural activity. Given recent successes using high-contrast “pings” to reveal hidden working-memory states, the authors hypothesize that similar pings could reveal a learned, latent spatial priority landscape without relying on task-related stimuli. The core research question is whether neutral visual pings during the intertrial interval can uncover the otherwise hidden, history-mediated spatial priority map and distinguish it from ongoing EEG activity lacking explicit signals of learned priority.

Prior work conceptualizes attentional control via priority maps integrating top-down goals, bottom-up salience, and selection history. Neuroimaging has largely focused on top-down/bottom-up effects, with limited evidence for sustained neural activity underlying selection history. Selection history effects (statistical learning, value-driven capture, intertrial priming) may arise via network-level synaptic plasticity, rendering them ‘activity-silent’ to conventional measures (ERP/BOLD). In working memory, pinging (high-contrast visual impulses or TMS) can reveal hidden memory contents by incidentally reactivating primed neural states. This suggests that latent structures, including statistically learned spatial priorities, might also be revealed by pings. Behavioral studies show rapid acquisition and persistence of spatial probability learning, and electrophysiological markers of top-down attention (e.g., alpha lateralization) do not readily explain selection history effects, motivating a ping-based approach to visualize history-modulated priority.

Participants: N=24 (17 female; mean age 24), normal or corrected vision, informed consent, ethics approved at Vrije Universiteit Amsterdam. Eight additional participants were replaced based on preregistered quality criteria (behavioral performance, RT outliers, EEG quality, fixation compliance). Task and Design: Modified additional singleton visual search with eight item positions on a circle. Targets (unique shape among distractors) required reporting line orientation. Salient color singleton distractors appeared on 70% of trials to increase task difficulty. Target location probability was biased within sequences of blocks: one high-probability (HP) location contained targets on 37.5% of trials (4.2× more likely than other locations). HP location cycled across four cardinal positions (up, down, left, right) across the session, with neutral blocks (equal probability across 8 locations) inserted to extinguish prior biases. Pings: On 50% of trials during the intertrial fixation period, a task-irrelevant high-contrast visual ping (200 ms) was presented 700–900 ms after fixation onset, with displays comprising four shapes (diamonds or circles) at the cardinal positions. Trials without pings included a trigger for time-matching/baselining only. Timing: Intertrial periods were time-matched across ping and no-ping trials. Search displays followed and remained until response or 2500 ms. EEG Acquisition: 64-channel BioSemi ActiveTwo, 512 Hz sampling, 10–10 layout; re-referenced to earlobes; VEOG/HEOG recorded. Preprocessing included high-pass filter at 0.01 Hz, epoching −700 to 1100 ms around ping onset (or matched timepoints for no-ping), artifact rejection with EMG-sensitive procedures and iterative interpolation, ICA (picard) for blink removal, bad channel interpolation, and exclusion of trials with saccades/drift via eye-tracking/EOG criteria. Eye-tracking: Eyelink 1000 (500–2000 Hz; downsampled to 500 Hz), blink interpolation, baseline in −200–0 ms pre-ping, trial matched to EEG. Eye-density and towardness analyses performed; dummy label control applied. Behavioral Processing: Accuracy and RT trimming per preregistration; exclusion of incorrect and extreme RT trials (~12% total). Decoding (MVPA): Cross-validated LDA using all 64 electrodes, trained separately for ping and no-ping trials. Classes: the four HP target locations (per block). Data baselined −200–0 ms relative to ping trigger (or matched timepoint in no-ping), downsampled to 128 Hz. To boost SNR (not preregistered), trial averaging (over 3 same-exemplar trials) and PCA (fit on training data within each fold) were applied; preregistered non-boosted analyses reported in Supplementary. Tenfold cross-validation with stratified folds; performance assessed by AUC over time from −200 to +600 ms relative to ping onset. Cluster-based permutation tests evaluated above-chance decoding and ping vs no-ping differences. Control Analyses: - Temporal confounds: Decoded the previous trial’s target location (8 classes; including neutral blocks) to remove block-phase structure; also applied a ‘dummy’ label scheme creating artificial temporal phases with matched trial counts—decoding collapsed to chance for dummy labels. - Intertrial priming vs learning: Re-ran HP decoding excluding trials where the prior trial’s target was at the current HP location (HP-exclusionary). Also an HP-only analysis trained on pings following HP-target trials. - Eye movement confounds: Eye-tracker based decoding and dummy analyses showed above-chance classification attributable to temporal noise; EEG decoding pattern did not mirror eye-based dummy decoding, ruling out eye confounds. ERP: Preregistered N2pc checks for search; lateralized ping responses assessed but not informative.

Behavior: Participants responded faster when targets appeared at HP locations than at low-probability locations (t(23)=10.62, p<0.001, dz=2.17, 95% CI [40.72, 60.48] ms). Intertrial priming: faster RTs when target location repeated vs switched (t(23)=5.956, p<0.001, dz=1.22, 95% CI [37.21, 76.82] ms). The HP advantage remained after excluding target location repetitions (t(23)=9.882, p<0.001, dz=2.02, 95% CI [29.56, 47.25]). Distractor interference was greater when the distractor appeared at the HP target location (t(23)=3.442, p=0.002, dz=0.70, 95% CI [10.41, 41.79]). Learning effects were consistent across phases (no interaction; BF10=0.073), and neutral blocks extinguished prior biases (t(23)=1.444, p=0.162). Excluding 11 participants who reported noticing regularities, the HP advantage persisted (t(12)=6.751, p<0.001, dz=1.90, 95% CI [25.27, 52.76]). Decoding: Robust above-chance decoding of the current block’s HP location emerged only following visual pings; no-ping trials showed little to no information about HP location. Cluster-based permutation tests confirmed ping decoding was above chance and significantly higher than no-ping. Temporal confounds ruled out: Decoding of the previous trial’s target location (8 classes) also appeared selectively after ping onset, and dummy-phase decoding collapsed to chance for both ping and no-ping trials. Intertrial vs statistical learning: HP-exclusionary analysis (removing trials preceded by HP-targets) still showed significant ping-evoked decoding of the HP location. HP-only analyses (training on pings following HP-target trials) also yielded strong decoding despite fewer trials, suggesting both intertrial priming and longer-term statistical learning contribute to the ping-revealed priority landscape. Eye movements: Eye-based decoding exhibited above-chance performance even with dummy labels, indicating temporal noise; EEG decoding did not, ruling out systematic gaze shifts as the source of ping decoding. ERP: No compelling lateralized ping-evoked component; search-related ERP checks behaved as expected.

The findings directly address whether latent, history-mediated attentional biases can be visualized. Despite no reliable evidence in ongoing EEG of the current HP location, neutral visual pings evoked patterns from which the HP location could be decoded, consistent with an activity-silent priority map mediated by synaptic-weight changes. Control analyses ruled out temporal structure and oculomotor confounds. The ping appears to incidentally mis-activate task-primed neurons whose activation thresholds are lowered by selection history, yielding weak but decodable spatial patterns. Behaviorally, both rapid intertrial priming and longer-term statistical learning shaped performance; neurally, ping decoding recovered both short-timescale (previous trial) and longer-timescale (blockwise HP) biases. These results support models positing synaptic mechanisms for selection history and demonstrate that history-based biases influence preparatory priority before stimulus onset, yet become apparent upon sensory drive (ping or probe). The work extends pinging beyond working memory, offering a new tool to study latent attentional states and integrating selection history into dynamic priority map theories.

This study introduces a neutral visual pinging approach to reveal the latent, history-modulated spatial attentional priority map using EEG and MVPA. It demonstrates that selection history biases, invisible in ongoing EEG, become decodable following brief sensory perturbations. Behaviorally, participants showed robust HP facilitation and enhanced distractor interference at HP locations; neurally, ping-evoked activity decoded both intertrial and statistically learned biases, independent of temporal and oculomotor confounds. These results suggest synaptic, activity-silent mechanisms underlie selection history’s influence on attention. Future work should combine pinging with neurophysiology to pinpoint mechanisms, test generalization to other selection history domains (e.g., reward history), probe feature-based priority maps, and refine impulse designs to dissociate proactive tuning from stimulus-evoked reactivation.

While consistent with activity-silent mechanisms, the approach cannot definitively exclude low-amplitude ongoing activity due to baselining/filtering. The blocked HP design poses potential temporal confounds, though multiple controls (previous-target decoding, dummy labels) mitigate this concern. The precise neural mechanism by which pings reveal latent priority remains unresolved. Eye movement confounds were carefully addressed, but subtle residual oculomotor influences cannot be entirely ruled out. Decoding does not uniquely specify representational formats, and SNR-boosting steps (trial averaging, PCA) were not preregistered (though preregistered analyses confirm the pattern).