Place sequence learning for navigation

Abstract

A model of the hippocampus as a “cognitive map”, inspired by the models of Burgess et al. (1994) and Jensen et al. (1996), is proposed. Simulations show that the resulting navigation behavior is as efficient as the behavior exhibited by previous models. However, the architecture of the proposed model and the mechanisms governing the temporal characteristics of the neurons in the model are more realistic. In particular, the proposed model assigns distinct and specific roles to the entorhinal cortex, the dentate gyrus and the hippocampal CA3 region. In essence, the dentate gyrus could act as a short-term memory that maintains the representation of a sequence of recently visited places. It could then force the corresponding CA3 place cells to fire and to learn the spatial relationships between places through a Hebbian rule. This “topological representation” could then serve as a basis for predicting places ahead of the animal and drive “goal cells”, i.e. cells that represent the direction to the goal, as proposed by Burgess et al.