Abstract Cognitive maps refer to internal representations of spatial and non-spatial relationships between entities (people or things) or events in the external world. There has been widespread excitement generated by recent discoveries that even continuous non-spatial task dimensions may be organized and ‘navigated’ as a cognitive map. These studies suggest the neural representations and computations revealed in physical space may be only one instance of a general mechanism for organizing and “navigating” any behaviorally-relevant continuous task dimensions (e.g. space, time, sound frequency, metric length). This insight raises the intriguing possibility that the well-established coding principles revealed during spatial navigation can also be used to understand flexible decision making in abstract and discrete tasks that are commonplace in the real world when they are based on a cognitive map, such as whom to collaborate with or where to eat. A cognitive map of an environment or task is incredibly powerful because it enables inferences to be made from limited experiences that can dramatically accelerate new learning and even guide novel decisions never faced before. Moreover, similar tasks that share an overall structure can be directly related to one another, thereby facilitating rapid generalization from one task or entity to another. Despite this wide-ranging importance for flexible cognition, we have only a basic understanding of how cognitive maps enable such novel inferences and generalization. Better understanding the mechanisms involved also carry significant clinical implications. Indeed, abnormal inferences, cognitive flexibility, and generalization are thought to core dysfunctions in several psychiatric conditions, ranging from schizophrenia to obsessive compulsive disorder to certain expressions of mood disorder. It follows that developing a mechanistic model of these component processes in humans has the potential to inform principled investigations into biomarkers and treatment targets for these disorders. The goal of this proposal is to develop a new neural model of how cognitive maps of abstract and discrete tasks are represented neurally and used to guide novel inferences during decision making in the human brain. We have developed a new experimental paradigm that induces people to form abstract and discrete cognitive maps (e.g. of social networks) and perform novel inferences during decision making. To develop our model, we will combine computational models of learning and inference in this paradigm with geometric models of neural coding derived from spatial navigation and “representational” and computational functional magnetic resonance imaging analysis methods that allow inferences to be made about the information represented and computations performed in different brain structures, respectively. The insights gained from this research will lead to substantial theoretical advances in models of goal-directed decision making...