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

Our visual world's complexity challenges our cognitive capacity. The brain simplifies perception through two principles: the predictability of repetitive environments (statistical learning) and the selective filtering of irrelevant information (attention). Statistical learning allows the brain to learn environmental regularities, sharpening perception around robust predictions and reducing computational costs. Selective attention prioritizes relevant information while suppressing irrelevant details, reducing perceptual redundancy. These mechanisms are interconnected; past experiences tune the visual system to expect relevant stimuli in specific spatial and temporal locations, integrating attention and statistical learning to optimize performance. Selection history, encompassing statistical learning and other history-based effects like value-driven attentional capture or intertrial priming, influences current behavior in attention. Selection history, along with top-down and bottom-up attentional mechanisms, converges in the attentional priority map—a real-time representation of stimulus behavioral relevance and saliency. This map, often studied in spatial features, codes attentional priority as weights in a topographic representation of space, with selection awarded to the highest-activity region. Selection history up-regulates weights to locations containing relevant past information and down-regulates those with frequent distractors. While priority maps are central to attentional selection theories, neuroimaging research has focused on top-down or bottom-up influences, with limited investigation into selection history's neural substrates. Preliminary studies suggest selection history effects are not based on sustained neural processes but rather on network-level synaptic plasticity. This latent characteristic makes selection history invisible to common neuroimaging techniques. The study of selection history effects has relied on evoked response differences between predictable and unpredictable stimuli as a proxy for latent expectations. A novel approach in working memory research uses high-contrast visual 'pings' to visualize activity-silent neural structures, suggesting that the ping technique might be applicable to visualizing learned attentional priority.

The literature review extensively discusses the existing research on statistical learning, selective attention, and their interaction. It highlights the concept of attentional priority maps and the limited understanding of selection history's neural mechanisms. The review emphasizes the challenge of studying activity-silent cognitive mechanisms, which are invisible to standard neuroimaging techniques, and introduces the novel 'pinging' technique used to visualize latent neural structures in working memory research. The authors argue that this technique might be adapted to visualize learned attentional priority, given the similarities between latent neural structures involved in working memory and statistically learned attentional priority.

The study used a modified additional singleton task with imbalanced target distributions. Participants (N=24) performed visual searches in biased blocks where targets appeared with higher probability at specific locations, intermixed with neutral blocks to reset learned priorities. Half the trials included task-irrelevant high-contrast visual pings during the intertrial period. Behavioral data (reaction times) were analyzed to assess the impact of target probability and location on performance. EEG data (64 electrodes) were recorded, and a classifier was trained to decode the high-probability target location from the EEG response to the pings, separately for ping and no-ping trials. Control analyses addressed potential confounds like temporal correlations and eye movements. Eye-tracking data were also collected and analyzed to rule out eye movement artifacts affecting the EEG decoding. Data preprocessing involved high-pass filtering, epoch creation, EMG artifact removal (using an adapted Fieldtrip procedure), ICA for eye-blink artifact removal, interpolation of bad electrodes, and trial rejection based on saccades or eye drift detected by eye-tracking. Multivariate pattern analysis (MVPA) using linear discriminant analysis (LDA) was applied to the EEG data, with the four high-probability locations as classes. Cluster-based permutation tests assessed the significance of decoding. An ERP analysis investigated lateralized evoked components (N2pc). Behavioral data analysis involved t-tests and ANOVAs; decoding analysis used cluster-based permutation tests and paired t-tests.

Behavioral results showed faster responses to targets at high-probability locations, a reliable intertrial target location effect, and slower responses when distractors appeared at high-probability locations. The effect of high-probability locations remained significant even after removing trials with repeated target locations. The neutral blocks successfully extinguished the attentional bias. Decoding analysis revealed robust above-chance decoding of the high-probability target location from ping-evoked activity but not from ongoing EEG activity in the absence of pings. Control analyses ruled out temporal confounds and eye movements as explanations for the decoding. An additional unplanned control analysis showed that the ping could also decode the preceding trial's target location. Excluding trials where the target was in the high-probability location on the preceding trial, high-probability target decoding remained significant, demonstrating that decoding was not solely due to intertrial priming effects. An exploratory analysis showed similar high decoding levels even when only trials with preceding high-probability target locations were used. This indicated that statistical learning across trials and intertrial priming both contributed to the observed decoding. Eye-tracking analyses demonstrated no significant relation between eye position and EEG decoding.

The findings support an activity-silent model of learned attentional bias. The ability to decode high-probability locations only with pings demonstrates that learned attentional priority is encoded in a latent layer of the attentional priority map, mediated by synaptic weight changes. The study demonstrates the ping technique's potential to visualize latent attentional biases. The fast updating of spatial priority maps highlights the flexibility of spatial probability learning. Although intertrial priming was present, the control analyses showed that the observed decoding reflected both intertrial effects and statistical learning. The results suggest that both intertrial effects and statistical learning might be mediated by synaptic mechanisms, and that 'pinging' could be a valuable tool for studying statistical learning and intertrial effects. The debate over why pings reveal hidden states focuses on whether the ping reactivates latent memories or simply reduces signal variance, revealing previously subthreshold activity. In this study, given the lack of sustained neural activity associated with selection history, the former interpretation seems more plausible. The study also touches upon the question of when learned expectations exert their influence (pre-stimulus vs. in response to sensory stimulation), suggesting that while statistical learning proactively adjusts the spatial priority map, this only becomes apparent after sensory input integration. The study concludes by emphasizing that the ping technique could be extended to study other latent cognitive biases beyond spatial attention.

This study successfully visualized the latent attentional priority map mediated by selection history using a novel 'pinging' technique. The results support activity-silent models of learned attentional biases and suggest that both statistical learning and intertrial priming effects are mediated by synaptic mechanisms. The pinging method offers a valuable tool for future research on latent cognitive biases in various domains.

One limitation is the potential for participants to have some awareness of the target probability manipulation, although this was mitigated by control analyses. The study's design does not definitively rule out the possibility of some subthreshold neural activity contributing to the observed effects. Further research is needed to fully clarify the underlying neural mechanisms responsible for ping-evoked decoding and the precise way in which visual pings reveal otherwise silent priority structures.