Mona Garvert: Stimulus-reward learning and generalization in structured environments

Max Planck Institute for Human Cognitive and Brain Science Leipzig

 

Abstract

The brain organises knowledge about the relationships between positions in space as well as non-spatial regularities in cognitive maps. Such a representation of knowledge is useful, because it can facilitate goal-directed behavior by enabling the generalization of value or other information across related states. Here, we combine a virtual reality task with computational modeling and functional magnetic resonance imaging (fMRI) to investigate how reward values that were never directly experienced can be inferred based on underlying structural regularities between stimuli. In this task, spatial relationships between stimuli learned on day 1 predict reward relationships in a choice task on day 2. We find that participants not only update the stimulus-reward associations they experience directly, but they also use their knowledge about the relationships between stimuli to predict values of stimuli which were not directly sampled.Similar behavior can be observed in a task where the underlying structural relationships between stimuli are non-spatial. Relational knowledge organized in cognitive maps can thus be used to extrapolate across related states and thereby facilitate novel inference.

 

Organized by

Lara Wieland/Lisa Velenosi