A Neurocomputational Approach to Trained and Transitive Relations in Equivalence Classes.

A stimulus class can be composed of perceptually different but functionally equivalent stimuli. The relations between the stimuli that are grouped in a class can be learned or derived from other stimulus relations. If stimulus A is equivalent to B, and B is equivalent to C, then the equivalence between A and C can be derived without explicit training. In this work we propose, with a neurocomputational model, a basic learning mechanism for the formation of equivalence. We also describe how the relatedness between the members of an equivalence class is developed for both trained and derived stimulus relations. Three classic studies on stimulus equivalence are simulated covering typical and atypical populations as well as nodal distance effects. This model shows a mechanism by which certain stimulus associations are selectively strengthened even when they are not co-presented in the environment. This model links the field of equivalence classes to accounts of Hebbian learning and categorization, and points to the pertinence of modeling stimulus equivalence to explore the effect of variations in training protocols.