Skip to main content
. 2019 Mar 18;29(6):979–990.e4. doi: 10.1016/j.cub.2019.01.077

Figure 1.

Figure 1

Grid Cell-Based Vector Computations and Visual Recognition Memory

(A) Grid cells in medial entorhinal cortex (MEC) exhibit periodic, hexagonally arranged firing fields, originally characterized as spatially selective cells in rodent experiments. Right: Spiking locations (red dots) superimposed on a rodent’s trajectory during foraging; Bottom left: A stereotypical, smoothed firing rate map.

(B) Vectorial coding in one dimension. 1D grid cells with appropriate phases across modules (top: 4 cells with different scales S1–4) project to distance cells. Distance cell arrays (2 for each direction along the 1D axis) project with monotonically increasing weights (W) to two readout cells (for up and down, respectively). The difference in output between readout cells is a measure of the distance and direction between start (blue) and goal (green) locations.

(C) Replicating distance cells and readout cells for a second, non-co-linear axis allows computation of 2D vectors in the stimulus (e.g., between facial features, to guide eye movements, e.g., in D).

(D) A face with superimposed saccade trajectories.

(E) Model schematic: Grayscale images are sampled by a square fovea (blue square). Feature detectors drive feature label cells, each coding for a particular salient feature. During training each feature label cell has been associated with a grid cell population vector (current position grid cells, blue cell and dashed arrows). All feature label cells of a given stimulus are bi-directionally connected to a single cell coding for the identity of the attended stimulus. Upon firing of an identity cell the currently active feature label cell is inhibited and identity cells select the next feature label cell to be active (short green arrow and cell), which is associated with its own grid cell population vector (target grid cells). Current and target position grid cell representations yield the next saccade vector (red arrow on image). Note, the selection of the feature label cell corresponds to a prediction of the next sensory discrimination (see also Figure S1). Image credit: Mr. Spock: public domain image; Grid cell rate map and rodent trajectory adapted from Barry and Bush Neural Systems & Circuits 2012 2:6, Attribution 2.0 Generic Creative Commons CC-BY 3.0; Saccade trajectories reproduces from Wikimedia Commons; Attribution 2.0 Generic Creative Commons CC-BY 3.0).