Place cells in the hippocampus are neurons with interesting properties such as the correlation between their activity and the animal’s position in space. It is believed that these properties can be for the most part understood by collective behaviours of models of interacting simplified neurons. Statistical mechanics provides tools permitting to study these collective behaviours, both analytically and numerically.
Here, we address how these tools can be used to understand place-cell activity within the attractor neural network paradigm, a theory for memory. We first propose a model for place cells in which the formation of a localized bump of activity is accounted for by attractor dynamics. Several aspects of the collective properties of this model are studied. The second part of this thesis deals with decoding place-cell activity. We introduce a decoding method based on the inference of an effective network and present its results on the processing of experimental recordings of place cells in a freely behaving rat.