Conclusions
With the direct input of place cells to grid cells and with the plasticity rule previously described, spatially correlated place cells compete to form the nodes of the grid map. As the animal explores the environment, some place cells are selected over the others to build the grid field. After approximately 30 minutes of exploring randomly an unfamiliar 1 meter square maze, the grid map emerges (see Figure 1). The gridness scores for the resulting grid maps are above 1. The synapses mediating this mechanism can represent the connections between cells in CA1 and the deep layers of mEC. An important feature of this model is that the formation of grid fields is not unique to spatial information: a regular firing rate lattice can be formed from an input population with competing neurons tuned for non-spatial information. Overall, our model is in close agreement with the recently discovered grid cells in primates' visual system [4] and in bat's cortex [5].