Theoretical and Computational Neuroscience
Author: Facundo Emina | Email: facuemina@gmail.com
Facundo Emina1°2°, Emilio Kropff2°
1° Universidad de Buenos Aires, Facultad de Ciencias Exactas y Naturales, Departamento de Física, Buenos Aires, Argentina
2° Fundación Instituto Leloir — IIBBA/CONICET, Buenos Aires, Argentina
Path integration is the ability that enables mammals to track their trajectory through space and navigate efficiently, even in the absence of external landmarks. In rodents, the entorhinal cortex contains various neurons that respond to navigational variables like angle and velocity, essential for this process. Within the same area, Grid Cells are thought to integrate this information, coherently firing with the animal’s position in space, forming a periodic lattice with triangular symmetry. However, the precise mechanisms behind this code remain unclear. Models based on continuous attractor neural networks suggest that grid cells achieve their firing pattern by integrating velocity and head direction signals within a specifically tuned recurrent network. Conversely, feed-forward models explain grid cell formation through self-organizing principles but do not account for path integration.
In this work, we explore how feed-forward models could be adapted to support path integration. Through numerical simulations, we show how adaptive and competitive networks could learn attractors from a tutor using Hebbian plasticity rules. Our goal is to provide a unified perspective on path integration by elucidating how these learned low-dimensional manifolds could facilitate effective spatial navigation.