T&C Chen Center for Social and Decision Neuroscience Seminar

Abstract: Cognitive maps are powerful internal representations of spatial and non-spatial relationships in the world that theoretically afford extreme forms of behavioral flexibility. In my talk I will present a series of studies showing that the hippocampus (HC), entorhinal cortex (EC), and orbitofrontal cortex (OFC) reconstruct subjective cognitive maps of abstract, non-spatial relationships sampled from separate experiences. In the first part, I will present work using geometric computational models across different tasks and species to show that novel direct inferences (zero-shot inferences) made over abstract cognitive maps use a grid code in EC and medial prefrontal cortex (mPFC) when inferences are composed on the fly during decision making tasks, ranging from choosing a social partner to making risky value-based choices. Collectively, this line of work suggests a new framework for understanding how values are computed and compared for flexible decisions. In the second part, I will present findings from a series of studies investigating how the brain uses an internal model of a task's structure, a type of cognitive map, to assign credit for reinforcement to latent (hidden) causes that support inference during learning. Using a Bayesian inference model that best explains learning behavior across species in similarly structured tasks, we find convergent evidence across humans and rodents for the encoding of inference-weighted prediction errors in phasic dopaminergic responses, supporting credit assignment to inferred associations. In tandem, the HC and OFC track the inferred "position" in the model-estimated association "space", recasting aspects of associative learning as a cognitive mapping problem. Collectively, these insights support theories proposing the brain tracks one's "position" in a cognitive map of an abstracted task space to afford efficient inferences for both learning and decision making. I will close by sharing my future research program in which computational modeling operates hand-in-hand with experimental design, simulation, and analysis and, through collaboration, promotes bi-directional translation across species.