Why grid cells function as a metric for space (Q6079060)

The aim in this article is to theoretically prove that in case of a large number of grid cells, the grid cell population codes can be taken as a metric for space. One considers an enhanced grid cell model based on Moiré grids. The activities of simple cells are simulated by trigonometric functions $e^{i\omega'}x$ and one defines $g:R^2\rightarrow C$ the function modelling a grid cell, the grid cell population codes, and the concept of metric as a positive definite kernel. In the Appendix part one proves that the reproducing kernel can be defined based on grid cells. One shows that the inner product of grid cell population codes converges exponentially to a positive definite kernel. An estimator for the true position is constructed. The parameters of the grid cells are trained in path integration tasks by maximizing a certain estimation. For the classification task a neural network with a recurrent layer, a linear hidden layer and an output layer is considered. Importance and application of grid cell population codes concept is outlined and described in the fourth section dedicated to grid cell population code navigation and fifth section where cell population codes in image classification are discussed. The sixth section of the article provides details and results on different experiments as: path integration, metric and path planning, navigation via grid population code, image classification via grid cell code. The last two sections are devoted to a general discussion on achievements and conclusions for future work.