Proceedings of the Annual Meeting of the Cognitive Science Society

Category learning is our ability to generalize across experiences and apply existing knowledge to new situations. Many real-world categories adhere to a “rule-plus-exceptions” structure, wherein most items are rule-followers, but a subset of “exceptions” violate category rules. Rule-plus-exception learning seems tightly coupled with hippocampal function. Though past work has demonstrated that prediction error drives hippocampus to form distinct representations of exceptions, limited work has investigated how this process impacts existing rule-follower representations. Here we use a neural network model of hippocampus to quantify how rule- follower representations are altered by the introduction of exceptions. By recording model representations of rule- followers before and after exceptions are introduced, we computed the shift in rule-follower representation elicited by exceptions. A rule-follower’s similarity to exceptions along category-relevant, but not irrelevant, dimensions predicted its degree of representational shift. This work furthers our understanding of how hippocampus supports the integration of surprising information in dynamic environments.