Home
This Title All WIREs
WIREs RSS Feed
How to cite this WIREs title:
WIREs Cogn Sci
Impact Factor: 2.218

Clarifying cognitive control and the controllable connectome

Full article on Wiley Online Library:   HTML PDF

Can't access this content? Tell your librarian.

Cognitive control researchers aim to describe the processes that support adaptive cognition to achieve specific goals. Control theorists consider how to influence the state of systems to reach certain user‐defined goals. In brain networks, some conceptual and lexical similarities between cognitive control and control theory offer appealing avenues for scientific discovery. However, these opportunities also come with the risk of conceptual confusion. Here, I suggest that each field of inquiry continues to produce novel and distinct insights. Then, I describe opportunities for synergistic research at the intersection of these subdisciplines with a critical stance that reduces the risk of conceptual confusion. Through this exercise, we can observe that both cognitive neuroscience and systems engineering have much to contribute to cognitive control research in human brain networks.

This article is categorized under:

  • Neuroscience > Cognition
  • Computer Science > Neural Networks
  • Neuroscience > Clinical Neuroscience
The brain and behavior in cognitive control (CC) research. (a) CC is recruited as a function of the expected value of its engagement computed by the dACC, control influences from the prefrontal cortex (PFC) and attention switching mechanisms in the posterior parietal cortex (PPC). (b) Behavioral costs of CC are often computed as the difference (e.g., in median differences across types of trials or differences on a specific trial) in response times between a “control‐demanding” experimental condition (teal bars) and a reference condition (blue bars). For example, this difference could represent a control‐demanding “interference effect” during a Stroop task or “switch cost” during a set‐switching task. The behavioral cost is thought to result from the push–pull between increasing task demands offset by recruited cognitive control (Botvinick et al., ). Accordingly, we can anticipate that influencing the brain's processing of task demands or core cognitive control circuitry will change the measured behavioral costs
[ Normal View | Magnified View ]
Node controllability at a high level in diffusion imaging. Network controllability analysis can identify the putative control roles of individual nodes in a network, such as has been applied to networks constructed from diffusion‐weighted imaging. Schematically, (a) within the brain, (b, left) anatomical regions can be represented in a parcellation combined with an estimate of the white matter pathways (e.g., “streamlines,” fractional anisotropy) of the brain reconstructed from diffusion imaging (b, right). (c) The streamlines can be used to construct a weighted region × region graph, or representation of the number of streamlines. (d) The location and identity of the regions and streamlines is retained in the original brain volume, facilitating (e) node controllability analysis to identify regions' characteristic control roles. AC, average controller; BC, boundary controller; MC, modal controller. Note that each node will exhibit some degree of controllability; here, maximally high ranking and distinct controllers are schematically represented
[ Normal View | Magnified View ]
Network controllability and energy landscapes. Top: A schematic energy landscape where each cell represents a possible state s N of the network. Peaks represent states with high energy. Moving from cell to cell constitutes a trajectory (t N) along the landscape. Moving farther or up an incline indicates a higher cost trajectory. Yellow cells represent special states where activity is integrated or segregated among all possible states. Bottom: In response to an input u N to nodes in different positions on the network given an arbitrary initial state s0, the network can be driven along distinct trajectories. Input u1 to node v1 will drive the brain into distant, hard to reach states (e.g., s1 via trajectory t1) and is thus high in modal controllability. Input u2 to node v2 will drive the brain into integrated or segregated states across clusters (e.g., s2 via trajectory t2), and is thus high in boundary controllability. Input u3 to node v3 will drive the brain along trajectory t2 into nearby, easy to reach states (e.g., s3 via trajectory t3), and is thus high in average controllability. The relationship between a neural state and its associated cognitive representation(s) and process(es) must be empirically determined at the appropriate neural scales. We can study how the brain naturally deploys CC to navigate the landscape as well as how to design control inputs to guide the mind
[ Normal View | Magnified View ]
A basic question in cognitive control research. From this toy example, we can observe that a fundamental control problem in an attention switch is to move the state of a perceptual system from a state visually encoding a house (left) to a state encoding a tree (right), such as with a switch of attention associated with an oculomotor saccade. The blue arrows represent the control process, which influences the relevant perception‐action network from one state to the other. These control processes could represent naturally occurring cognitive control, cognitive control moderated by exogenous stimulation, or stimulation alone. The details of the controller is what we aim to “reverse engineer” from this initial intuition
[ Normal View | Magnified View ]
A schematic diagram for the natural “controller/controllee” duality in neural control engineering at an arbitrary neural scale. (a) At any given location in the brain, such as the prefrontal cortex (a), we can consider the local organization of (b) white matter (which comprises the anatomical connections between brain regions) and gray matter (which comprises the neural cell bodies which sum neural inputs and initiate outputs). Some proportion of white matter connections relay signals into the gray matter (yellow connections) and some relay signals out of the gray matter (blue). (c) At a smaller scale, gray matter is generally arranged into layers that include numerous different types of cells organized within layers (represented with colored symbols) which communicate among each other. Certain layers ultimately project information out of the local layers to other regions, representing the net outputs of computations that occur within gray matter that will influence other brain regions. In this case, the pictured brain region is the “controller” that influences other regions. For simplicity, the interlayer interactions are not pictured here, but crucial to the computations in real networks. (d) Exogenous brain stimulation can be applied to local brain regions (now the “controllee”), modulating the typical neural activity across layers (here represented by the green arrows, and ultimately the output of the brain to other regions
[ Normal View | Magnified View ]
Goals for control theory in cognitive control research. In principle, we can focus on two relationships when linking control theory to cognitive control research. Schematically, (a, left) we should consider the relationship between control inputs and resulting neural activity. (a, right) We should also identify the relationships between neural activity and cognition, often measured using behavioral variables in cognitive control research—the lynchpin of cognitive neuroscience. Because neural activity joins the panels in (a), the system identification aspect of control theory will involve a focus on identifying equations that express neural dynamics' natural and manipulable relationships with cognitive control behavioral output. (b) Notably, we can also be interested in effects directly from inputs to behavioral outputs in control theory. In each graph, the characteristic function will be associated with a noise profile, which limits our ability to uncover and exploit the basic input–output curves
[ Normal View | Magnified View ]
Neural states and outputs in cognition. In an arbitrary perceptual example, we can imagine an initial state s0 of a set of neurons that can exist in several other states (s1, s2, s3) that result when the system is influenced by a Control Input (e.g., cognitive control or exogenous stimulation, including environmental cues or invasive or noninvasive neuromodulation such as via brain stimulation). However, only s1 is associated with a tree, whereas s2 and s3 are associated with a house. If a person perceives a house and we cannot directly measure the internal (neural) state, we cannot know which state the system is in. However, we could identify an input that results in the perception of a house, and the perception of the house is more “robust” because, all else being equal, if the system cannot achieve s3, then s2 is still a viable state to achieve to ensure that the person will perceive the house
[ Normal View | Magnified View ]
A schematic diagram of a Watt governor. James Watt introduced an early flyball governor in 1788 to control the velocity of steam engines. It contained both a sensor and a control mechanism. (a) A connecting belt from the engine spun a cylindrical shaft. (b) As the engine's speed increased, the centrifugal force created by the shaft's rotation spun the flyballs. (c) As the flyballs elevated from a resting state s0 to a higher state st with increasing engine velocity, a mechanism transferred the flyball displacement along a stem connected to a steam valve. (d) As engine speed increased, a steam valve opened, reducing engine pressure and velocity
[ Normal View | Magnified View ]

Browse by Topic

Computer Science > Neural Networks
Neuroscience > Clinical Neuroscience
Neuroscience > Cognition

Access to this WIREs title is by subscription only.

Recommend to Your
Librarian Now!

The latest WIREs articles in your inbox

Sign Up for Article Alerts