Proceedings of the Annual Meeting of the Cognitive Science Society

We present a neural process model of visual analogical mapping that receives image inputs and responds by spatially selecting a matching object to a cued object. The relational structure of the base scene is stored in a way that specifies the arguments of each relation, allowing mappings based on structural correspondence to be represented as proposed by the structure-mapping theory. All the processes in the model emerge out of coupled integro-differential equations modeling neural population activation dynamics. The mapping can be influenced by both featural and relational similarities. The developmental shift in mappings in the presence of a featural distractor can be accounted for by manipulating how well the model can maintain attention to relevant feature/relation dimensions, consistent with a hypothesis suggesting inhibitory as a key factor explaining the shift.