Efficient coding in biophysically realistic excitatory-inhibitory spiking networks.

Area of Science:

• Computational neuroscience
• Neural network modeling
• Information theory

Background:

• The principle of efficient coding suggests neural networks maximize information transmission with minimal energy.
• It's unclear if this principle alone explains empirical neural activity properties.

Purpose of the Study:

• To derive structural, coding, and biophysical properties of spiking neural networks based on efficient coding principles.
• To investigate if efficient coding can explain fundamental properties of neural network activity.

Main Methods:

• Derived network properties by minimizing an instantaneous loss function and a time-averaged performance measure for efficient coding.
• Modeled excitatory-inhibitory recurrent networks of spiking neurons encoding stimulus features.
• Assumed stimulus features varied at the timescale of neuronal membrane constants.

Main Results:

• The optimal network exhibited biologically plausible features: integrate-and-fire dynamics, spike-triggered adaptation, and external excitatory input.
• Excitatory-inhibitory recurrent connectivity with similar tuning implemented feature competition, enhancing coding efficiency.
• Network properties, including neuron ratios and connectivity, matched biological cortical networks.
• Achieved instantaneous excitation-inhibition balance and efficient coding across multiple timescales.

Conclusions:

• Efficient coding, by minimizing cost and maximizing information, can account for key structural, coding, and biophysical properties of biological neural networks.
• This normative principle provides a unifying framework for understanding neural computation.