BEHAVIORAL EVIDENCE FOR A GENERAL COGNITIVE MAP: INFERENCE IN A MULTIDIMENSIONAL ABSTRACT STIMULUS SPACE

We designed a behavior to address questions related to how rats maintain representations of a complex stimulus space, often described as a cognitive map. We rewarded select stimuli from a two-dimensional auditory space composed of one hundred different stimuli. Experimental design was focused on disentangling stimulus-response based learning strategies, as exemplified by Reinforcement Learning (RL) theory from that of a map-like representation of a stimulus space. We found that rats can create configural, flexible representations of an abstract space and use this information to infer the presence of rewards in novel situations. These findings are inconsistent with most model-free RL strategies but are consistent with map-like strategies for representing an abstract stimulus space. This behavior provides a powerful approach for further exploring the applicability of neural representations of physical space to abstract spaces