Neural Dynamics of Numerical Cognition: Coding, Maintenance, and Interference in the Primate Frontoparietal Network

Abstract

This dissertation investigates how the brain of non-human primates encodes, retains and recognizes numerical information, combining behavioral and neuronal approaches. Understanding how numerical cognition emerges from distributed neural processing offers a unique opportunity to explore the interaction between working memory, recognition memory and decision-making – three core components of higher cognition.

Electrophysiological recordings from the dorsolateral prefrontal (dlPFC) and posterior parietal (PPC) cortices of rhesus monkeys performing a sequential delayed match-to-numerosity task revealed that single neurons encode recognition memory for numerical quantities. During comparisons of the sample stimulus with up to three test stimuli, the neurons exhibited distinct yet interacting patterns of familiarity modulation, including repetition suppression and match enhancement. These results suggest that recognition memory encompasses abstract, non-symbolic representations, and that the ability to detect familiarity is distributed across prefrontal and parietal networks.

At the population level, the interaction between dlPFC and PPC was further characterized by analyzing simultaneously recorded neuronal activity using canonical correlation analysis (CCA). This approach revealed dynamic, direction-specific coupling between both regions that varied across task phases. During stimulus encoding, information flow was predominantly feedforward from parietal to prefrontal areas, whereas maintenance periods were marked by more balanced interactions. These results suggest that numerical recognition arises from the flexible coordination of the frontoparietal network as a whole, rather than from the isolated processing of a single brain region.

Extending these findings to a comparative perspective, behavioral performance in the same task was examined across species, revealing that rhesus monkeys and carrion crows show comparable accuracy and numerical effects, such as distance and size effects. Despite these similarities, the temporal dynamics of decision-making differed systematically between species. Monkeys exhibited increasing reaction times across successive test phases, whereas crows responded progressively faster. Machinelearning analyses further demonstrated that reaction times carried information about numerosity and task phase in monkeys but not in crows, suggesting species-specific strategies in numerical recognition and decision control.

Together, these studies provide a comprehensive, multi-level account of numerical cognition, covering from single-neuron encoding and network dynamics to behavior across species. By integrating neurophysiological and behavioral evidence, this thesis sheds light on abstract cognitive functions, such as number recognition.