Computational Model-Based Cognitive Neuroscience

We show how the functional resolution of mapping neural circuit features to distinct cognitive and behavioral components of decision-making process can be improved, by combining hierarchical latent-mixture and evidence-accumulation based models with neural (fMRI) data. These models are implemented within a Bayesian inference framework. Theory based and structural assumptions are used to develop evidence accumulation models whose parameters are hierarchically governed by cognitive subsystems, such as performance monitoring, belief updating, error-feedback, and executive control. Such models explicitly account for the temporal dynamics and learning associated with repeated decision making. They dissociate between different potential cognitive processes and strategies that may be used on a trial-by-trial basis, and account for the hierarchical process of strategy switching. The hierarchical cognitive processes and subsystems are characterized by cognitive parameters that potentially capture and mediate the relationship between neural (fMRI) and behavioral data. Such neuro-cognitive modeling allows us to differentiate between theories, provide insights into the developmental maturation of brain networks, and improve the identification of differential brain feature characteristics associated with different cognitive processes in clinical and neuro-diverse populations. Applications include identifying the joint neural and behavioral basis of individual differences in mathematical decision making, response inhibition, and perceptual decision making tasks.

To examine how the brain produces behavior, new statistical methods have linked neurophysiological measures directly to mechanisms of cognitive models, modeling both modalities simultaneously. However, current simultaneous modeling efforts are largely based on either correlational methods or on functions that map one stream of data to the other. Such frameworks are limited in their ability to infer causality between brain activity and behavior. We investigate one causal framework for explaining how behavior can be viewed as an emergent property of brain dynamics. Our proposed framework can be considered an extension of multivariate dynamical systems (MDS; Ryali et al. Neuroimage, 54(2), 807–823, 2011), as it is constructed with temporal dynamics and brain functional connectivities. To test the MDS framework, we formulate a concrete model, demonstrate that it generates reasonable predictions about both behavioral and fMRI data, and conduct a parameter recovery study. Specifically, we develop a generative model of perceptual decision-making in a visual motion-direction discrimination task. Two simulation studies under different experimental protocols illustrate that the MDS model can capture key characteristics of both behavioral and neural measures that typically occur in experimental data. We also examine whether or not such a complex system can be inferred from experimental data by evaluating whether current algorithms for fitting models to data can recover sensible parameter estimates. Our parameter recovery study suggests that the MDS parameters can be recovered using likelihood-free estimation techniques. Together, these results suggest that our MDS-based framework shows great promise for developing fully integrative models of brain-behavior relationships.

Characterizing the temporal dynamics of functional interactions between distributed brain regions is of fundamental importance for understanding human brain organization underlying higher-order cognition. Progress in the field has been hampered both by a lack of strong computational techniques to investigate brain dynamics and an inadequate focus on core brain systems involved in higher-order cognition. Here we address these gaps by developing a novel variational Bayesian Hidden Markov Model (VB-HMM) that uncovers non-stationary dynamical functional networks in human fMRI data. VB-HMM revealed multiple short-lived cognitive states characterized by rapid switching and transient connectivity between the salience (SN), default mode (DMN), and central executive (CEN) networks—three brain systems critical for higher-order cognition. Notably, in children, relative to adults, VB-HMM revealed immature dynamic interactions between SN, CEN, and DMN, characterized by higher mean lifetimes in individual states, reduced switching probability between states and less differentiated connectivity across states. Our findings suggest that the flexibility of switching between distinct cognitive states is weaker in childhood, and they provide a novel framework for modeling immature brain network organization in children. More generally, the approach used here may be useful for investigating brain dynamics in neurodevelopmental disorders with deficits in higher-order cognition.

To better understand human behavior, the emerging field of model-based cognitive neuroscience seeks to anchor psychological theory to the biological substrate from which behavior originates: the brain. Despite complex dynamics, many researchers in this field have demonstrated that fluctuations in brain activity can be related to fluctuations in components of cognitive models, which instantiate psychological theories. In this talk, I will describe a statistical framework (joint modeling) that links computational models of behavior to neuroimaging data by exploiting patterns of covariation between the streams of data. I will then describe a number of recent advancements that address issues pertaining to scalability, determining significance of brain-behavior connections, and continuous spatio-temporal dynamics