The representational geometry of hierarchical decision-making processes

About This Grant

RESEARCH SUMMARY Goal-directed behavior in dynamic, naturalistic environments depends on a hierarchy of decision-making processes that operate at different timescales to infer context, adjust decision policy, and map stimuli to actions. Still, the neural mechanisms underlying hierarchical decisions are poorly understood. To study these processes, we propose a project based on recordings from large neural populations in the lateral intraparietal area (LIP), supplementary eye field (SEF), and dorsolateral prefrontal cortex (dlPFC) of macaque monkeys performing a hierarchical decision task. This task is a variant of the direction discrimination task with a random dots stimulus in which monkeys classify the motion direction of the stimulus while also tracking a hidden, spontaneously changing environment variable. Preliminary analysis demonstrates that monkeys can perform the task by approximating a normative strategy that requires monkeys to track variables on multiple timescales: the monkey must integrate motion evidence to make individual decisions and must integrate decision feedback and confidence to infer the context over the span of many trials. To understand the representational and computational mechanisms that support these dynamic and temporally multiplexed codes, I hypothesize that hierarchical decision variables are flexibly represented in orthogonal subspaces in each population to enable parallel computations and generalization across conditions. To investigate this, I will study the representational geometry in each recorded region through linear decoding, dimensionality reduction, and manifold capacity analyses. Additionally, preliminary results also demonstrate that the representations of both perceptual decision variables (i.e., dot motion), and contextual decision variables (i.e., environment) are distributed across all recorded regions. I hypothesize that this distributed code is maintained through task-relevant communication subspaces, which I will characterize by identifying representational dependencies and signal latencies across regions. Ultimately, this project will use innovative analyses and a novel, well-controlled task to provide a comprehensive understanding of the neural mechanisms – both within regions and across regions – that underly hierarchical decision-making processes in the primate brain. Clarifying such mechanisms has broad applicability in our understanding of general cognition at large and may ultimately inform new diagnostic tools and therapeutic strategies for conditions that involve abnormalities in decision-making and belief-updating, such as depression, anxiety, and schizophrenia.