Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session A: Tuesday, August 12, 1:30 – 4:30 pm, de Brug & E‑Hall

Prefrontal Representations During Learning Reflect Probabilistic Computations Across Domains

Fahd Yazin¹, Gargi Majumdar, Neil R Bramley², Paul Hoffman; ¹University of Edinburgh, ²University of Edinburgh, University of Edinburgh

Presenter: Fahd Yazin

The prefrontal cortex (PFC) is thought to represent abstract forms of cognitive maps or internal models during tasks. These representations could be specialized structures suited for distinct domains of experience (e.g., people vs places). Alternatively, they could represent domain-general processes rather than structure, suited for inference across domains. Here we tested these competing accounts using a learning task where human participants learned probabilistic cognitive maps in an unsupervised manner, across three domains, while performing rule classifications. During spatial, social and sequential learning, we found that the structured 1D map representations are formed in the entorhinal cortex but not in midline PFC. Instead, the PFC performs probabilistic inference, abstracting out the underlying probability distributions. Specifically, the ventromedial PFC computes data likelihood under different models, updating them through experience akin to a Bayesian learner. The anteromedial and dorsomedial PFC represent (angular) directional changes and transition distances respectively, within this abstract probability space. These findings were seen during inference as well on unseen exemplars. These results suggest that the midline PFC might be performing a domain-general computation on learned cognitive maps - probabilistic search.

Topic Area: Predictive Processing & Cognitive Control

Extended Abstract: Full Text PDF