This research article explores the intricate relationship between recurrent network structure and the resulting low-dimensional neural activity. By integrating the theories of neural fields and low-rank networks, the authors provide a mathematical framework that links single-neuron selectivity to population-level dynamics on neural manifolds. Their findings reveal that while distinct circuit structures can unexpectedly produce identical latent dynamics, the underlying connectivity still leaves a detectable "footprint" on neuronal activity. These structural constraints manifest as specific symmetries in dynamics and topological patterns in similarity space. Ultimately, the study offers a theoretical bridge to help neuroscientists use recorded activity to infer the hidden functional organization of biological neural circuits.

References:

• Pezon L, Schmutz V, Gerstner W. Linking neural manifolds to circuit structure in recurrent networks[J]. bioRxiv, 2024: 2024.02. 28.582565.