Conference on Cognitive Computational Neuroscience

By representing multidimensional relationships in a map-like structure, cognitive maps enable the brain to infer new relationships and support flexible decision making. Previous research has implicated the entorhinal cortex (EC), hippocampus (HC), and medial prefrontal cortex (mPFC) in cognitive map formation and representation, but their respective contributions have been a matter of considerable debate. In particular, a fundamental question concerns to what extent these regions support context-dependent (specific and flexible) versus context-invariant (abstract and generalizable) representations. We designed a novel ‘wine space’ task that asked participants to form and use a cognitive map of wines to flexibly select the best wine in several different market contexts. Univariate effects of the context-dependent rank difference (i.e. decision value) were identified in the mPFC and HC/EC, consistent with flexible use of task-relevant relational values for decisions. Furthermore, whole-brain representational similarity analysis (RSA) of wine rank differences revealed an abstraction hierarchy, with EC showing a generalized code, the HC an axis-specific code, and the mPFC tracking only contextually-relevant information. These results suggest that the brain employs multiple regions to track cognitive map relationships at multiple levels of abstraction, and recruits contextually relevant information to compute decision variables to meet task demands.