After a general introduction and a brief review of the available experimental data on spatial representations (chapter 2), this thesis is divided into two main parts. The first part, comprising the chapters from 3 to 6, is dedicated to grid cells. In chapter 3 we present and discuss the various models proposed for explaining grid cells formation. In chapter 4 and 5 we study our model of grid cells generation based on adaptation in the case of non-planar environments, namely in the case of a spherical environment and of three-dimensional space. In chapter 6 we propose a variant of the model where the alignment of the grid axes is induced through reciprocal inhibition, and we suggest that that the inhibitory connections obtained during this learning process can be used to implement a continuous attractor in mEC.
The second part, comprising chapters from 7 to 10 is instead focused on place cell representations. In chapter 7 we analyze the differences between place cells and grid cells in terms on information content, in chapter 8 we describe the properties of attractor dynamics in our model of the Ca3 net- work, and in the following chapter we study the effects of theta oscillations on network dynamics. Finally, in Chapter 10 we analyze to what extent the learning of a new representation, can preserve the topology and the exact metric of physical space.
