Abstract
Navigation in the world uses a combination of visual, path integration, and planning mechanisms. Although visual cues can modify and stabilize navigational estimates, path integration signals provide “ground truth” upon which vision builds, and enables navigation and dead reckoning in the dark. A complete understanding of visually-based navigation thus requires an understanding of how path integration creates the spatial representations upon which vision can build. Grid cells in the medial entorhinal cortex use vestibular path integration inputs to generate remarkable hexagonal activity patterns during spatial navigation (Hafting et al., 2005). Furthermore, there exists a gradient of grid cell spatial scales along the dorsomedial-ventrolateral axis of entorhinal cortex. It has been shown how a self-organizing map can convert the firing patterns across multiple scales of grid cells into hippocampal place cell firing fields that are capable of spatial representation on a much larger scale (Gorchetchnikov and Grossberg, 2007). Can grid cell firing fields themselves arise through a self-organizing map process, thereby providing a unity of mechanism underlying the emergence of entorhinal-hippocampal spatial representations? A self-organizing map model has been developed that shows how path integration signals may be converted through learning into the observed hexagonal grid cell activity patterns across multiple spatial scales. Such a model overcomes key problems of the useful oscillatory interference model of grid cell firing (Burgess et al., 2007). The proposed new model hereby clarifies how path integration signals generate hippocampal place cells through a hierarchy of self-organizing maps. Top-down attentional matching mechanisms are needed to stabilize learning in self-organizing maps (Grossberg, 1976). Such hippocampal-to-entorhinal feedback mechanisms illustrate how visual cues can build upon and modify entorhinal and hippocampal spatial representations during navigation in the light.
Supported in part by CELEST, an NSF Science of Learning Center (SBE-0354378) and the SyNAPSE program of DARPA (HR00ll-09-3-0001, HR0011-09-C-0011).