Constructing neural network models from brain data reveals representational transformations underlying adaptive behavior

Abstract

The human ability to adaptively implement a wide variety of tasks is thought to emerge from the dynamic transformation of cognitive information. We hypothesized that these transformations are implemented via conjunctive representations in conjunction hubs – brain regions that selectively integrate sensory, cognitive, and motor representations. We used recent advances in using functional connectivity to map the flow of activity between brain regions to construct a task-performing neural network model from fMRI data during a cognitive control task. We verified the importance of conjunction hubs in cognitive computations by simulating neural activity flow over this empirically-estimated functional connectivity model. These simulations produced above-chance task performance (motor responses) by integrating sensory and task rule information in conjunction hubs. These findings reveal the role of conjunction hubs in supporting flexible cognitive computations, while demonstrating the feasibility of using empirically-estimated neural network models to gain insight into cognitive computations in the human brain.

Causally informed activity flow models provide mechanistic insight into network-generated cognitive activations

“…As mentioned above, another important component in the ideal study would be to follow the causal chain from stimulus to the cognitive phenomenon of interest (i.e., 2-back vs. 0-back activity increases in right DLPFC) to motor responses (behavior). This more comprehensive explanation of the DLPFC n-back effect is beyond the scope of the present study, but something close to this level of explanation-of a different set of neurocognitive phenomena-has been achieved in a recent actflow study (Ito et al, 2021). Finally, the ideal study would use stimulation and lesion approaches to fully verify the causal relevance of observed neural signals.…”

Network modeling of dynamic brain interactions predicts emergence of neural information that supports human cognitive behavior

“…The above output features (fully dynamic and prospective predictions) bring the activity flow framework into the realm of artificial neural network models (ANNs) that have emerged as the dominant method of simulating cognitive information in artificial intelligence (Ito et al, 2020; Yamins and DiCarlo, 2016; Yang and Wang, 2020). A critical benefit of our approach over these artificial models is its foundation in empirical data, which enabled 1) empirical estimation of network connectivity weights and 2) direct comparison of model-derived representational geometry with actual empirical data.…”

“…Future extensions of our modeling approach might also target the role of the CCNs in the emergence of information in the more formal sense, by disambiguating when/how new information emerges in the brain, versus charting the spread of information after its initial emergence. This will likely require refinement of dynamic activity flow modeling to integrate non-linear operators (Ito et al, 2020; Rigotti et al, 2013).…”

Generalization in Sensorimotor Networks Configured with Natural Language Instructions

“…One influential systems-level explanation of our ability to generalize performance posits that flexible inter-region connectivity in the pre-frontal cortex (PFC) allows for the reuse of practiced sensorimotor representations in novel settings [1, 2]. By recombining practiced stimulus-response patterns according to the ostensible demands of a previously unseen task, we can leverage well-established abilities and perform on new tasks in very few practice trials.…”