The widely investigated phenomenon of place cells found in rodent hippocampus offers a huge selection of features to be modeled. In this project we concentrate on the bell curved shape of the place cell activity within the cell's place field, and how this activity profile can be learned from preprocessed sensory input perceived during exploratory motion. We build a neural network model of the hippocampal CA3 region composed of units of simplified pyramidal neurons.

Our hypothesis is that it is not necessary to assume the afferent input to be smoothly changing in space. That is, spatially proximal locations (and orientations) of the rat may evoke completely different afferent excitation patterns on the CA3 region. However the spatiotemporal continuity may provide a temporal correlation between input patterns originating from locations close in space. In order to be able to use the traditional correlation based learning rules a mechanism must be elaborated which sustains the activities of pyramidal neurons after the corresponding input has become inavailable (due to the continuous motion of the rat). In other words the network must show a slowly decaying short term memory capability. The densely interconnected recurrent projections of CA3 provide an ideal substrate to implement this type of memory by self-exciting reverberant loops.

One important motivation of our work is to find a connection between the different functional roles attributed to hippocampus, namely the (episodic) memory function (mostly referred in context of primate studies) and the "cognitive map" function (observed in rodents). The paradigm of temporal correlation based learning may bridge the conceptual gap between these two aspects of hippocampal function.

Máté Lengyel and Zoltán Szatmáry were working on this project.