Grid-cell models for navigation and context discrimination

TitleGrid-cell models for navigation and context discrimination
Publication TypeConference Paper
Year of Publication2009
AuthorsKubie, J. L., Fenton A. A., Lytton WW., & Burgess N.
Conference NameSociety for Neuroscience 2009 (SFN '09)
KeywordsSFN, Society for Neuroscience
Abstract

Although entorhinal grid cells are assumed to be part of an animal's navigational machinery, there are few proposed mechanisms. We propose models based on the recent finding that grid cells are organized in groups from dorsal to ventral that we will call ``grid spacing groups''. While the grid spacing for cells within a region is fixed, spacing increases in discrete steps from dorsal to ventral in each entorhinal cortex (Barry et al, Nat. Neuro. 10, 2007). Grid cells within a group have the property of ``fixed spatial relations'' such that relationship between neighboring bumps from a pair of grid cells is fixed both within and across environments (Fyhn et al. Nature 446, 2007). For example, if grid cells A and B are in the same spacing group and one of B's bumps is slightly NW of the nearest A bump, then this relationship will be constant for all neighboring bumps within the environment and across all environments. We define a firing vector as the set of cells that are active at a moment in time. The navigation model relies on the assumption that, during a lifetime of learning, the animal associates a short movement vector, such as a short movement NW, with a predicted transition of active grid cells within a group. Firing vector 1 plus a short movement vector leads to firing vector 2. For the cells of a grid spacing group the number of firing vectors and translations is small. A goal is represented as the firing vector, across all grid spacing groups, at the goal location. From the animal's current location (and firing vector) an exhaustive search model has the animal evaluate linear journeys. Evaluating the path in a single direction involves a series of small movements in a that direction. For each movement the firing vector is updated by the ``fixed spatial relations'' rule. This continues either until the goal-vector is matched or the search distance is exhausted. This pattern is repeated, in series, in all directions until the goal match is found. The context discrimination model relies on the assumption that the property of ``fixed spatial relations'' applies within a spacing group, but not across spacing groups. Consider groups from the right and the left entorhinal cortex with the same grid spacing. All pairs of cells from these two groups would have fixed spatial relations within one environment, with the the relations between bilateral pairs maintained by sensory cues. Across environments, with no sensory constraints, the cells of the EC on each side could translate independently. Bilateral cell pairs would shift their spatial relations. The result of an extreme change in the environment would be independent, non-overlapping sets of firing vectors. Each firing vector set would represent a context.