Flexible Use of Limited Resources for Sequence Working Memory in Macaque Prefrontal Cortex

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• The brain's working memory (WM) capacity is limited, yet it can generalize to novel information.
• Mechanisms governing the rational allocation of limited WM resources are not well understood.

Purpose of the Study:

• To investigate how the prefrontal cortex allocates working memory resources during sequence working memory (SWM) tasks.
• To elucidate the neural mechanisms underlying flexible resource utilization and generalization in WM.

Main Methods:

• Utilized calcium imaging and electrophysiological recordings in non-human primates performing SWM tasks.
• Analyzed neural representations, including geometric structures and subspace dynamics, in relation to WM load.

Main Results:

• Neural representations of SWM, including geometric and rank subspaces, dynamically changed with WM load.
• Working memory resources, quantified by neural signal strength and tuning, were flexibly shared across SWM ranks.
• The prefrontal cortex employed distinct neural strategies: shared tuning for generalization and disjoint tuning to minimize interference, achieving a cost-benefit trade-off.

Conclusions:

• The geometry of compositionality is fundamental to the flexible use of limited working memory resources.
• Neural resource allocation in the prefrontal cortex dynamically predicts behavioral outcomes in SWM tasks.