Help Privacy Policy Disclaimer
  Advanced SearchBrowse





Distinct excitatory and inhibitory connectivity structures control the dynamics and computational capabilities of recurrent networks

There are no MPG-Authors in the publication available
Fulltext (restricted access)
There are currently no full texts shared for your IP range.
Fulltext (public)
There are no public fulltexts stored in PuRe
Supplementary Material (public)
There is no public supplementary material available

Giannakakis, E., Buendia, V., Vinogradov, O., Khajehabdollahi, S., & Levina, A. (2024). Distinct excitatory and inhibitory connectivity structures control the dynamics and computational capabilities of recurrent networks. Poster presented at Computational and Systems Neuroscience Meeting (COSYNE 2024), Lisboa, Portugal.

Cite as: https://hdl.handle.net/21.11116/0000-000E-6FB2-D
Connectivity among cortical pyramidal neurons is far from random. Notably, it is known that there is a higher connection probability between similarly tuned neurons across cortical layers. In theoretical studies, this type of connectivity has been shown to generate rich dynamics in spiking networks. While the connectivity of inhibitory neurons is usually less specific than that of excitatory neurons, several studies have found different degrees of inhibitory specificity and diverse connectivity patterns, whose impact on population dynamics and computational implications are not fully understood. Here we link the presence of inhomogeneous localised, neuron-type specific clusters with complex dynamics that are associated with optimal performance in computational tasks. Using a reservoir computing framework, we evaluate the computational capabilities of balanced, recurrent net- works of rate neurons with neuron-type specific connectivity patterns. The reservoirs are composed of several interconnected clusters of E/I populations and are trained to simultaneously predict the trajectories of multiple chaotic time series. We study the impact of varying E and I clustering levels on network dynamics and identify the optimal topology for a complex time series reconstruction task. We find that the presence of different levels of excitatory and inhibitory clustering enables the precise control of the network’s dynamical state. In particular, we show that E and I clustering levels distinct and non-trivial effects on network dynamics and identify structures that control the network’s distance from the chaotic state. Finally, we demonstrate that a commonly observed cortical connectivity pattern of highly specific excitation and less specific (but not uniform) inhibition among similarly tuned neurons can maintain network dynamics close to the edge of chaos and may significantly contribute to the computational efficiency of brain networks.