Abstract

Grid cells in the entorhinal cortex of the mammalian brain are thought to provide a multi-scale periodic representation that functions as a metric for coding space. Years of research have been focusing on uncovering the mechanisms and functional significance of this spatial representation in navigation. Many recent works have found that training neural networks to perform navigation tasks in 2D arenas based on velocity inputs can lead to the emergence of grid-like spatial response patterns, under some biologically plausible constraints. Why and under what conditions grid cell patterns emerge from such navigation-optimised neural networks remains controversial.

We will first give an introduction to grid cells and place cells, followed by path integration tasks solved by recurrent networks where grid-like representations are observed and are beneficial for goal finding in unfamiliar and changing environments. We than present an analytic theory supporting grid-like firing fields through the lens of pattern formation, and then opposing views about the suspected brittleness of the phenomenon, and its dependence on post-hoc implementational choices.

References:

[1] Andrea Banino, Caswell Barry, Benigno Uria, Charles Blundell, Timothy Lillicrap, Piotr Mirowski, Alexander Pritzel, Martin J Chadwick, Thomas Degris, Joseph Modayil, and others. Vector-based navigation using grid-like representations in artificial agents. Nature, 557(7705):429, 2018

[2] Sorscher, Ben, et al. A unified theory for the computational and mechanistic origins of grid cells. Neuron, 2022

[3] Schaeffer R, Khona M, Fiete I. No free lunch from deep learning in neuroscience: a case study through models of the entorhinal-hippocampal circuit. Advances in Neural Information Processing Systems, 2022

[4] Ben Sorscher, Gabriel C. Mel, Aran Nayebi, Lisa Giocomo, Daniel Yamins, and Surya Ganguli. When and why grid cells appear or not in trained path integrators. bioRxiv, 2022

Zoom information:

https://eng-cam.zoom.us/j/84204498431?pwd=Um1oU284b1YxWThObGw4ZU9XZitWdz09 Meeting ID: 842 0449 8431 Passcode: 684140