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

Poster Session B: Wednesday, August 13, 1:00 – 4:00 pm, de Brug & E‑Hall

Value GradientRescales Grid-like Representations during Reward Learning

Hai-Tao Wu¹, Lusha Zhu¹; ¹Peking University

Presenter: Hai-Tao Wu

Cognitive maps are hypothesized to organize abstract features spatially, akin to physical environments, to guide decision and generalization. We propose that to better facilitate reward learning, these maps may distort by how fast value changes (i.e., value gradients), over-representing dimensions where feature changes yield larger reward differences. Using fMRI, we tested this during a reinforcement learning task with jellyfish images varying in two features (spot number, tentacle number) as options. Participants learned value maps where one feature had twice the reward sensitivity of the other. Under the assumption of internal maps scaled by value gradients, we observed six-fold periodic BOLD signals—a signature of grid-like coding—in entorhinal cortex (EC) and medial prefrontal cortex (mPFC). Compared with a range of alternative scales, signal strength peaked around value gradient scale and correlated with choice accuracy. These results point to a possibility that cognitive maps may be optimally constructed to reflect the reward structure in service of goal-directed behavior.

Topic Area: Reward, Value & Social Decision Making

Extended Abstract: Full Text PDF