Fig. 1. Leveraging the Guided Activation Theory to inspire ENN models of cognitive computation during task-based fMRI.

a A modified version of the Guided Activation Theory of prefrontal cortex, highlighting a potential key role for conjunction hubs. The Guided Activation Theory posits that sensory cortices (left), which contain sensory stimulus-related activations, and prefrontal areas (top), which contain task context/rule activations, integrate in association cortex to produce conjunctive activations through patterns of guided activations. Conjunctive activations are then guided to motor areas to generate motor response activations for task behavior. b The Guided Activation Theory can be reconceptualized in a connectionist framework. This provides a formalization of how flexible sensorimotor transformations may be implemented computationally. The formalization involves the task context and sensory stimuli representing the input layer, the association units representing a hidden layer, and the behavioral (motor) responses as the output layer. c Testing the Guided Activation Theory using task fMRI data collected in humans during context-dependent tasks. Using quantitative methods, we empirically test how different task activations (e.g., sensory stimuli and task context) form conjunctive activations to produce motor response activations using activity flow mapping¹⁸. d The Guided Activation Theory can be empirically tested by projecting task activation patterns between brain areas by estimating inter-area FC weight mappings obtained from resting-state fMRI data. Based on the activity flow principle¹⁸, we estimated inter-vertex mappings using regression (see Methods) on resting-state fMRI data. This approach identifies a projection that maps across distinct spatial units (i.e., vertices) in empirical data, similar to how inter-layer weights propagate activity across layers in an ANN.