Abstract
Abstract
Cognitive effort is closely linked to prefrontal control, but the control demands imposed by working memory remain poorly understood at the level of neural circuits. We formulated frontoparietal regulation as a feedback-control problem in a conductance-based spiking network comprising an intraparietal ring attractor and a recurrent dorsolateral prefrontal cortex (dlPFC) actuator. A local data-driven task–actuator model was identified around the cue-offset mnemonic state and used to design linear-quadratic regulation. Increasing memory load produced a strong monotonic rise in realized post-cue control effort, driven primarily by sustained dlPFC recruitment. Cue-aligned selective routing supported higher mnemonic loads with substantially less control effort than spatially nonspecific routing. Mean-matched open-loop stimulation reproduced much of the nominal retention benefit, whereas state-dependent feedback provided greater robustness when mnemonic activity was perturbed. Finite-horizon analysis revealed a strongly anisotropic, effectively low-dimensional control geometry dominated by a common mnemonic direction, with much weaker access to item-differentiating directions. These results link increasing working-memory demand to greater sustained prefrontal control requirements and show how frontoparietal routing geometry shapes the efficiency and robustness of mnemonic regulation.