Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster E in Poster Session E: Thursday, August 6, 10:30 am – 12:15 pm, Kimmel Center, Shorin & Rosenthal Rooms

Topographic constraints drive the emergence of grid-cell center-surround connectivity in path-integrating networks

Navneet Prakash¹, Mayukh Deb², Kushal Reddy Dudipala³, Apurva Ratan Murty³; ¹Yale University, ²Google, ³Georgia Institute of Technology

Presenter: Navneet Prakash

Grid cells in the medial entorhinal cortex (MEC) fire in periodic, hexagonal patterns and are wired with a characteristic center-surround recurrent connectivity. Prior computational models have reproduced some of these features, but typically by hand-designing the relevant wiring or by recovering structure after post hoc neuron sorting. Neither of these strategies tells us *why* this pattern exists and whether such a connectivity can emerge directly through learning. Here we trained a recurrent neural network to perform path integration while encouraging nearby neurons to have similar outgoing connectivity. This constraint introduces spatial geometry over neurons and breaks the permutation symmetry, thereby biasing the network toward spatially organized solutions. We find that this model not only performs path integration effectively and develops grid-like representations, but also exhibits clear center-surround recurrent connectivity *without* any sorting step. These findings suggest that center-surround wiring in the MEC may emerge naturally from the interaction between task demands and topographic constraints, rather than requiring specially engineered wiring rules.

Topic Area: Methods, Tools, Theory & Neural Coding