CAREER: Neural circuit mechanisms of inference: how brains learn and use hidden structure

New York University, New York NY

Investigators

Abstract

Remarkably, animals can exhibit sophisticated behaviors in novel environments, despite having limited experience with them. How does the brain make inferences about the underlying statistics and generative structure of environments, and use those inferences to guide behavior? While specific brain regions have been implicated in this capacity, it is unclear how patterns of neural connections and activity in these regions mediate inference and sophisticated reasoning. This project delineates the neural circuit mechanisms spanning multiple brain regions by which animals make inferences and use those inferences to guide their behavior. The research group of this study has recently developed a high-throughput behavioral training facility for rats that trains dozens of rats (~60) per day on sophisticated cognitive behaviors. The group leverages this high-throughput approach to relate individual differences in rats’ strategies to differences in neural connectivity and neural activity recorded during behavior. In parallel, the principal investigator develops a workshop for high school students in New York City that applies lessons from neuroscience, psychology, and economics to improve financial literacy and decision-making. Financial literacy provides a unique outreach opportunity for increasing awareness of neuroscience research and communicating cutting edge neuroscientific results to the broader public, in the context of a highly salient and relatable topic: personal finance. While the research project seeks to understand the neural mechanisms by which brains make inferences to guide decision-making, the educational activities synergistically seek to educate students about the neuroscience behind their economic choices, and empower them to make better ones. This project examines whether and elucidates how neurons in the lateral orbitofrontal cortex (lOFC) make synaptic connections with genetically-defined cell types in the dorsomedial striatum (DMS) in service of model-based reinforcement learning. Rats are trained on a novel task with hidden structure that, if inferred, provides a clear behavioral read-out of model-based reasoning. Viral tracing, multi-regional paired recordings during behavior, and projection-specific recordings and inactivations are deployed to identify the dynamics and connectivity of subcircuits within the lOFC that generate model-based inferences. These experiments are performed in dozens of rats, and leverage individual differences in behavior to identify projection-specific prefrontal subnetworks that subserve model-based reinforcement learning. The project provides a circuit- and synaptic-level understanding of a core computation, that of an economic reference point against which outcomes are evaluated, compared, and chosen, in rats that infer reference points according to an internal model of the task. The results of this work will have broad implications for the fields of neuroscience, psychology, and machine learning. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.