CNS 2018: Learning Relative Landmark Locations

Click on image to enlarge

This poster walks through a proposal of how the brain uses grid cells to perform unsupervised learning of landmark locations. It shows results of a network model trained on 1000 environments, each with 16 locations containing random landmarks from a pool of 5 unique landmarks. The network is able to distinguish between environments with substantial noise, while a bag-of-features model is not.

Background

  • Grid cells provide location codes, spatially related through path integration
  • Multiple grid cell modules provide unique location codes for many large environments
  • We have shown that displacement modules encode spatial relationships between grid cell reference frames

Contributions

  • We propose that displacement cells encode the relative position of pairs of landmarks and that sets of displacement cells provide robust representations of environments
  • Simulations show the model’s ability to learn and distinguish among many complex environments with high noise tolerance