Winnerless competition in clustered balanced networks: inhibitory assemblies do the trick.

Area of Science:

• Computational Neuroscience
• Neural Network Modeling
• Mammalian Neocortex Dynamics

Background:

• Balanced networks are fundamental models of mammalian neocortical activity, characterized by canceling excitatory and inhibitory inputs.
• Existing models extended to include clustered excitatory neurons explain persistent activity but face limitations with firing rate saturation.
• Cortical working memory tasks exhibit high trial-to-trial variability and reduction during stimulation, posing challenges for current models.

Purpose of the Study:

• To review mean field descriptions of balanced networks and their application to clustered topologies.
• To address the firing rate saturation issue in clustered balanced networks inconsistent with experimental data.
• To propose and analyze a novel network architecture with locally balanced excitatory and inhibitory clusters.

Main Methods:

• Review of mean field theory for balanced networks of binary neurons.
• Application of mean field theory to clustered excitatory connectivity.
• Development and analysis of a novel model with locally balanced excitatory and inhibitory neuron populations.
• Validation through numerical network simulations.

Main Results:

• Stable fixed points in networks with purely clustered excitatory connectivity rapidly approach firing rate saturation.
• The proposed model with locally balanced excitatory and inhibitory clusters demonstrates true multistability.
• Activated clusters in the novel model exhibit moderate firing rates across a broad parameter range.

Conclusions:

• The standard clustered balanced network model's tendency towards firing rate saturation is a significant limitation.
• Locally balanced joint excitatory and inhibitory clustering offers a promising framework for realistic neural network modeling.
• This novel approach supports multistability and moderate firing rates, better reflecting experimental observations in cortical networks.