Neural & Behavioral Models

Pedestrian road-crossing is a safety-critical decision in which people transform continuously evolving perceptual information into a binary choice (cross vs wait) and a response time. While single-trial joint models of neural and behavioral data are well established in mathematical psychology (Turner et al., 2013; Ghaderi-Kanghavari et al., 2023), they have rarely been pushed into naturalistic, dynamic decisions studied in transport and road-user behavior. Here we apply an integrative, single-trial EEG-behavior evidence accumulation model to a controlled pedestrian crossing experiment with repeated manipulations of time-to-arrival (TTA) and simultaneous EEG recording. The model treats choices, response times, and a trial-wise centro-parietal positivity (CPP) summary (O'Connell et al., 2012) as co-generated by shared latent accumulation dynamics, enabling formal tests of alternative neural commitments (e.g., CPP reflecting drift vs urgency/boundary changes). Because closed-form joint likelihoods are unavailable, we fit a hierarchical version of the model using amortized simulation-based Bayesian inference. Posterior predictive checks show that a single parameterization reproduces both the behavioral (choice/RT) patterns and the CPP feature distributions. The model also supports cross-modal prediction: early EEG windows sharpen predictions of imminent crossing choices and response times, and observed behavior yields predictive distributions over CPP on the same trial. Finally, trial-level coupling can be stress-tested via EEG-behavior trial shuffling, which degrades model fit as expected if alignment is meaningful. These results demonstrate a practical route to bringing single-trial neurocognitive modeling into pedestrian road-crossing research, and more broadly into dynamic, real-world decision settings.

This is an in-person presentation on July 20, 2026 (09:00 ~ 09:20 EDT).

In multisensory and within-sense interactions, there are four kinds of neurons which we modeled to address perceptual interactions in psychophysical data. We find: (1) Facilitatory bimodal binocular and multisensory neurons are well modeled using Minkowski’s (1910) nonEuclidian distance equation. In two cases where there were comparable psychophysical data, the parameterizations agreed for audiovisual neural and psychophysical data, but diverged for binocular neural and psychophysical data. (2) Some sensory neurons fire for only one kind of stimulus, but increase their firing rate if another stimulus is present. We modeled this behavior with a gated power law amplification. Eight sets of neural and psychophysical data from multisensory and within-sense (chromatic and binocular) interactions were beautifully fit by this model and yielded similar power law exponents. In two cases where we had comparable neural and psychophysical data (binocular neurons and binocular detection thresholds; audiovisual neurons and brightness-amplified-by-audio) the fitted power law exponents agree. (3) Suprathreshold binocular contrast follows E. Schrödinger’s (1926) nonlinear averaging model. We fit this model to four populations of binocular and multisensory suppressive neurons and to situationally suppressive visual-vestibular neurons. Schrödinger’s model mimics aspects of MLE reliability-weighted models for psychophysical combination. (4) Some neurons fire for only one kind of stimulus, but reduce their firing rate when a second kind of stimulus is present. We modeled this in binocular and audiovisual neurons with a gated power law suppression. Similar behaviors occur in psychophysics. Funded by a supplement to ONR MURI N00014-20-1-2163.

This is an in-person presentation on July 20, 2026 (09:20 ~ 09:40 EDT).

The brain expends energy in the performance of mental work but there have been limited attempts to empirically link measures of cognitive work with measures of energy expended. We previously suggested that the hazard function of the reaction time (RT) distribution, h(t), can be interpreted as an instantaneous measure of the amount of work being performed. We here suggest that the global field power (GFP) of electroencephalographic (EEG) data can be interpreted in terms of brain energy expenditure. Forming a ratio of h(t) and the GFP gives a ratio that we refer to as the neural efficiency score (NES). We used the RTs from trials on which correct responses were made to estimate h(t) and we used the EEG data from the same trials to calculate the GFP. We found that h(t) was reliably higher for iron-sufficient (IS) than for iron deficient non-anemic (IDNA) women but converged for the longest RTs, suggesting, IS women were accomplishing more work than IDNA women. This ordering was reversed in the GFP data, suggesting that IDNA women were expending more neural energy to accomplish their perceptual work than were IS women. Combining these variables in the NES showed that IS women had a higher level of neural efficiency than did IDNA women and this was true across the range of RTs.

This is an in-person presentation on July 20, 2026 (09:40 ~ 10:00 EDT).

Does the stop signal reaction time (SSRT) really measure the time it takes for the brain to inhibit a response? We will expose three fundamental problems with this common assumption. First, the SSRT is primarily made of visual and motor delays, which account for 40% of its individual differences. Second, SSRT is by design influenced by how often participants try to stop and, when calculated as a single measure, will therefore always over-estimate the time it takes to stop when they try. Last, the SSRT is estimated from behavioural reaction time, and therefore cannot directly inform on when a process takes place in the brain, unless motor execution time distribution is estimated. Relying on multiple visuo-manual stop task datasets, we will show how distributional analyses of reaction time can lead to robust and more meaningful temporal estimates of when key mechanisms start in the brain, including automatic and reactive inhibition.

This is an in-person presentation on July 20, 2026 (10:00 ~ 10:20 EDT).