Optimal Population Coding for Dynamic Input by Nonequilibrium Networks.

Area of Science:

• Computational Neuroscience
• Statistical Physics
• Information Theory

Background:

• The efficient coding hypothesis posits that neural responses maximize information transmission about external stimuli.
• Existing models often simplify network connectivity and stimulus dynamics, limiting biological relevance.
• Understanding neural encoding in complex, dynamic environments remains a challenge.

Purpose of the Study:

• To investigate optimal neural population coding in networks with asymmetric connectivity encoding dynamic stimuli.
• To characterize stimulus encoding strategies and network dynamics under varying input statistics.
• To bridge nonequilibrium physics with neural coding and network dynamics.

Main Methods:

• Utilized a kinetic Ising model to simulate neural population responses to dynamic input.
• Applied gradient-based methods and mean-field approximation for network reconstruction.
• Analyzed network asymmetry, decoding performance, and entropy production.

Main Results:

• Identified stimulus encoding strategies influenced by input correlation, timescale, and reliability.
• Demonstrated that network dynamics are altered by the statistics of dynamic input.
• Found an optimal effective temperature in asymmetric networks facilitating efficient coding.

Conclusions:

• Asymmetric network connectivity and dynamic input statistics significantly impact optimal neural coding strategies.
• The study provides a framework connecting nonequilibrium physics to neural dynamics and coding.
• This work advances our understanding of how biological neural networks encode complex, time-varying information.