Item

Abstract

Cognitive maps represent relational structures crucial for generalization and flexible decision-making in both spatial and non-spatial domains. Although their benefits have been explored extensively, the way they are learnt remains less clear. We introduce a graph-structured sequence task to investigate how cognitive maps are learned and adapted for generalization and decision-making. Participants learned about a sequential transition structure that led to one of two fluctuating rewards. The task design enabled generalization of value across three of four possible sequences, and importantly two of these three sequences were also perceptually similar. Behavioral data revealed participants’ ability to adapt flexibly to changes in reward and leverage the afforded generalization successfully. These findings provide insights into how cognitive maps are learned and suggest corresponding neural changes.