Neural Models

Event-related brain potentials that index perceptual, cognitive, and motor processes have been explained in biophysical terms, but few have been explained in computational terms. Neural correlates of visual attention during visual search have been described with an event-related potential of the electroencephalogram (EEG) known as the N2pc and with measures of neural discharges in the frontal eye field (FEF) and other cortical areas. We show that a neurocomputational model explaining the evolving salience representation of FEF visual neurons (Cox et al., 2022) provides a computational explanation of visual search mechanisms spanning neurophysiology, electrophysiology, and behavior. This model assumes that a salience representation is produced by interacting computations of detecting stimuli (localization) and matching each stimulus to a target (identification). Here we show that the timing and amplitude of the N2pc can be predicted by the dynamics of localization, identification, and salience representations describing FEF neurons sampled simultaneously. Crucially, mathematically opposing identification and salience signals are necessary to account for the variation of N2pc latency and amplitude with the number of distracting items. Such opposing signs could be a consequence of identification and salience being associated with different biophysical sources. These findings demonstrate the utility of rigorous, explainable cognitive models for translating across levels of description and scales of measurement and offer new insights into the neuro-computational origin of a commonly studied and clinically important EEG signal.

The brain maintains neural timelines of the remembered past and the anticipated future. Exponentially-decaying cells reach peak firing after an inciting event and come back to baseline firing with a continuous spectrum of time constants. These activities form a basis set for the past timeline and can be identified with Laplace transform of the past. In parallel, exponentially ramping cells (Cao, et al., 2024 PNAS) provide a basis set for the time of anticipated future events. We describe a cognitive model applied to behavioral timing tasks that uses this neural representation. In the model, temporal relationships between events are learned and stored in a Hebbian weight matrix associating the past to the present. By querying this neurally-realistic representation with task appropriate cues, the network outputs possible timelines that follow the cue. Coupled with straightforward decision-making models, this hybrid neurocognitive framework provides qualitative fits to a range of timing behavioral paradigms in animals and people, replicating canonical behavioral patterns including scalar property and regression-to-mean. By representing temporal relationships between events as functions, rather than a simple strength, the model is able to account for adaptive behaviors when the timing of reward is dynamic with high uncertainty. This ability is highlighted in tasks such as willingness-to-wait, where participants' temporal beliefs are systematically manipulated (McGuire & Kable 2012, Cognition).

The Late Positive Component (LPC) has consistently been observed as an Electroencephalography (EEG) correlate of recognition memory task performance and was suggested to be functionally related to memory recollection. Recently, the LPC was suggested to be potentially related to decision-making processes. Here, we investigated whether the LPC was associated with trial-level memory strength, as represented by the drift rate parameter in the diffusion decision model (DDM). Building on recent advancements, we modified the DDM to predict trial-level LPC amplitude, decision accuracy, and reaction time while accounting for EEG measurement noise. Neural networks (using BayesFlow) were trained to obtain parameter estimates for the models. Data were collected from 24 participants over 3 EEG recording sessions with a recognition memory task. Our results demonstrated the model's ability to account for both behavioural and neural observations across experimental conditions. By comparing the variability in LPC amplitude explained by across-trial drift rate variability, we established a link between LPC and trial-level drift rate. Furthermore, the selectivity of this association was demonstrated by showing minimal variance in LPC accounted by non-decision time variability. Overall, compared to previous studies of LPC, we provided a stronger test for the functional role of LPC in representing memory strength variability.

One can instantly identify a dog or grab a cup without much effort. This ease in navigating the world is the result of countless interactions with the environment. While automatic behaviors typically allow for reaching goals with little cognitive effort expenditure, some automatic behaviors are problematic and lead to difficulties that can prevent achieving long-term goals. In contrast, cognitive control is effortful, but it can allow for halting undesirable automatic behaviors. In this experiment, we used electrophysiology to test the relationship between well-known cognitive control measures and button-switch interference in a well-practiced categorization task. Updating scores were negatively correlated with accuracy button-switch interference while a negative correlation between stop signal reaction time and button-switch RT interference was trending. With regards to EEG, the feedback-locked P2 and P3a were larger for negative feedback than positive feedback trials during the button-switch categorization session. Recordings during the SST showed a trending negative correlation between the motor inhibition independent component and SSRT. Lastly, stimulus-locked microstate (MST) analysis during the button-switch categorization task identified four MST. Higher updating scores were related to spending more time in MST3 (sourced in the left insula), while the frequency of occurrence of MST1 (sourced in the right angular gyrus) was related to accuracy button-switch interference. The correlation between SSRT and probability of transition between these two microstates was also trending. Together, these results suggest that individual differences in cognitive control can account for differences in the inhibition of automatic behaviors, and that EEG can be used to identify biomarkers.