Individualized prediction of human pattern detection under conditions of “unknown unknowns”

Functional neuroimaging has provided insights into which regions or connections between regions facilitate various cognitive functions in humans. But the insight comes only after averaging across a cohort of subjects. There is still no individualized diagnostic and assessment tool based on functional MRI for clinical neuroscience.  We fill this gap by moving towards individual level fMRI prediction in the context of studying how humans handle uncertainties beyond the "risky vs ambiguous" scenario, in a task design presenting humans with uncertainty about uncertainties and probabilities as feedback.  We employ ultra-high field (7T) MRI imaging in combination with a hierarchical Bayesian model of oscillator network and timeseries prediction approaches towards prediction of individual subject's "state of mind" along the "confident vs. confused" gradient as extracted from fMRI.  For the purposes of probing how humans make decisions under severe and unpredictably changing levels of ambiguities with a feedback that is unreliable, and given as probabilities, we designed a cybersecurity-themed self-paced game task that requires subjects to detect patterns from a sequence of decision and feedback cycles spanning multiple trials.

Cybersecurity Game: In real life, feedback is ambiguous. Here, individuals learn complex patterns that indicate threat. The task was developed in conjunction with intelligence analysts as a model for real-world detection of cybersecurity threats.

Work currently in progress; stay tuned for a preprint!