A Control-Theoretic Account of Cognitive Effort in Working Memory

Working memory and cognitive effort are intimately connected to prefrontal control, yet the neural circuitry behind this regulation remains unclear. Researchers approached this problem by modeling frontoparietal regulation as a feedback-control issue within a spiking neural network featuring an intraparietal ring attractor and a recurrent dorsal lateral prefrontal cortex (dlPFC) actuator.

By identifying a local data-driven task-actuator model around the cue-offset mnemonic state, they were able to apply linear-quadratic regulation. As memory load increased, there was a steady, monotonic rise in post-cue control effort, largely due to persistent dlPFC activation. Selective routing aligned with the cue facilitated higher mnemonic loads with less control effort compared to routing that was spatially nonspecific.

Mean-matched open-loop stimulation replicated many of the benefits related to retention, but state-dependent feedback proved more robust when mnemonic activity was disrupted. Finite-horizon analysis indicated a highly anisotropic, effectively low-dimensional control geometry, primarily driven by a common mnemonic direction, with less access to item-specific routes.

These findings establish a clear link between heightened working-memory demands and increased sustained prefrontal control requirements, while also demonstrating how the specific routing geometry within the frontoparietal system influences the efficiency and stability of mnemonic regulation.