Efficient coding in biophysically realistic excitatory-inhibitory spiking networks.

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Theoretical Neuroscience

Background:

• The efficient coding principle suggests sensory cortical networks maximize information transmission while minimizing metabolic energy expenditure.
• However, it remains debated whether empirical properties of neural network activity can be solely explained by this normative principle.

Purpose of the Study:

• To derive the structural, coding, and biophysical properties of spiking neural networks based on efficient coding principles.
• To investigate if minimizing an instantaneous loss function and a time-averaged performance measure can explain neural network characteristics.

Main Methods:

• Derived properties of excitatory-inhibitory recurrent networks of spiking neurons by imposing efficient coding constraints.
• Assumed encoding of independent stimulus features with time scales matching neuronal membrane time constants.
• Analyzed emergent network properties including dynamics, connectivity, and input characteristics.

Main Results:

• The optimal network exhibits biologically plausible features: integrate-and-fire dynamics, spike-triggered adaptation, and non-specific excitatory input.
• Excitatory-inhibitory recurrent connectivity with similar tuning implements feature competition, mirroring visual cortex findings.
• Optimal neuron ratios and connectivity patterns resemble those in cortical sensory networks, with instantaneous excitation-inhibition balance.

Conclusions:

• Efficient coding principles can account for fundamental structural, coding, and biophysical properties of biological neural networks.
• The derived network model demonstrates efficient coding capabilities even with stimuli varying across multiple time scales.
• These findings support efficient coding as a unifying framework for understanding neural network organization and function.