Computational Neuroscience Seminar
Connectivity Structure and Collective Dynamics in Neural Circuits
Speaker: David G. Clark, Kempner Institute, Harvard University
Location: Warren Weaver Hall 1314
Videoconference link: https://nyu.zoom.us/j/9726507138?omn=91057366818
Date: Tuesday, February 3, 2026, 4 p.m.
Synopsis:
Theoretical studies of neural-circuit dynamics have long relied on models with random, unstructured connectivity. While analytically tractable, this assumption is increasingly at odds with modern neuroscience data. Large-scale connectomics reconstructions, including whole-brain wiring diagrams in Drosophila, reveal that biological circuits exhibit rich spectral structure, with features including rapidly decaying singular-value spectra and structured overlaps between singular vectors that deviate dramatically from random-matrix predictions. Simultaneously, advances in high-density recording technologies enable measurement of collective activity across thousands of neurons, shifting focus from single-neuron responses toward population-level features like dimensionality, which are more directly tied to computation and behavior. Despite these parallel revolutions in measuring connectivity and activity, we lack analytical frameworks linking them: How does structured connectivity shape collective dynamics? I will present recent work addressing this question. First, I will discuss a theory of activity dimensionality in random networks with i.i.d. couplings (Clark et al., PRL 2023), showing that chaotic dynamics generate extensive yet low-dimensional activity with long collective timescales. I will then introduce the random-mode model (Clark et al., PRX 2025), a tractable connectivity ensemble capturing key spectral features of real connectomes. Studying networks with this connectivity, we show that structure can be invisible at the single-neuron level yet dramatically shape collective activity, with dimensionality depending on just two parameters: coupling variance and effective rank.