Population coding in sparsely connected networks of noisy neurons

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Neural Networks

Background:

• Sensory information encoding in the neocortex relies on coordinated neural populations due to individual neuron variability.
• Population coding is fundamentally linked to network structure, as neurons receive and transmit information within the network.

Purpose of the Study:

• To investigate the relationship between population coding and spatial connection statistics in simulated neural networks.
• To determine if sparse, local connectivity in a modeled cortical sheet can support robust information encoding.
• To identify structural requirements beyond random spatial probabilities for high-fidelity neural networks.

Main Methods:

• Modeled a sheet of cortical neurons with sparse, primarily local connections.
• Simulated network activity to assess encoding of internal state variables.
• Varied connection probabilities and network structures to evaluate their impact on signal-to-noise ratio and fidelity.

Main Results:

• A sparse, locally connected network model demonstrated the capacity to encode multiple internal state variables with high signal-to-noise ratio.
• Randomly instantiating connections based on spatial probabilities did not yield high-fidelity networks.
• Achieving high-fidelity encoding necessitated additional structural organization, including higher cluster factors and correlated inputs to neighboring neurons.

Conclusions:

• Sparse, local connectivity is a viable basis for population coding in neural networks.
• Network structure significantly influences the fidelity of population coding, beyond simple connection probabilities.
• Specific structural features, such as clustering and input correlations, are essential for high-performance neural information processing.