Tuesday, August 8 2023
11:30 - 12:30

Alladi Ramakrishnan Hall

Emergence of structure in cortical circuits through bottom-up dynamical principles

Sarthak Chandra

Massachusetts Institute of Technology, USA

Modularity and hierarchical organization are fundamental features in biology, but the mechanisms underlying their emergence during development remain elusive. While some information is likely genetically coded, development must unfurl through the dynamics of self-organization that obeys fundamental physics principles. In this talk, I will demonstrate how simple bottom-up dynamical rules can give rise to rich structure in the form of functional modularity and hierarchy.

I will focus on two key examples: grid cell modules in the entorhinal cortex, and the hierarchical organization of the visual cortex. For grid cells, I introduce the novel principle of peak selection, whereby local interactions organize module boundaries from a global gradient, unifying the positional hypothesis and the Turing pattern formation hypothesis for morphogenesis. I will show that peak selection in the entorhinal cortex results in the formation of robust grid modules through a form of topological quantization with invariance to all microscopic details. Further, we make robust predictions for the relationship between grid cell modules, yielding the most accurate match to grid cell data to date. In the visual system, I will show how a competitive synaptic growth rule driven by spontaneous spiral waves on the retinal surface can lead to the emergence of a spatially arranged visual cortical hierarchy with primate-like retinotopy. Our proposed model remarkably requires only a small number of parameters, and robustly reproduces several additional features of sensory cortices, such as the formation of mirrored maps, local convolutional-like connectivity and variation in receptive field size and structure through the visual field.

Through the formulation of such simple dynamical principles governing the development of cortical circuits, our models produce multiple testable hypotheses for future developmental, connectomics and physiology studies for sensory cortices and the entorhinal cortex.

Download as iCalendar