Large-Deviation Approach to Random Recurrent Neuronal Networks: Parameter Inference and Fluctuation-Induced...

Area of Science:

• Computational neuroscience
• Statistical physics

Background:

• Neuronal networks exhibit complex dynamics often analyzed using statistical mechanics.
• Large deviations theory provides tools to study rare events and fluctuations in stochastic systems.

Purpose of the Study:

• To unify field-theoretical methods with large deviations theory for analyzing neuronal networks.
• To develop a data-driven approach for parameter inference in recurrent neural networks.
• To investigate fluctuation-induced phenomena in neural dynamics.

Main Methods:

• Developed a field-theoretical framework for a prototypical random recurrent network model with continuous units.
• Derived the rate function using field theory, identifying it with the effective action.
• Utilized the Kullback-Leibler divergence form of the rate function for parameter inference.

Main Results:

• The effective action in the field-theoretical approach is identical to the rate function.
• The rate function is expressed as a Kullback-Leibler divergence, enabling parameter inference.
• Identified a regime where fluctuations induce transitions between mean-field solutions.

Conclusions:

• The unified framework provides a powerful tool for analyzing neuronal network dynamics and fluctuations.
• Data-driven inference of model parameters is facilitated by the derived rate function.
• Fluctuation-induced transitions represent a novel mechanism in neural dynamics beyond mean-field approximations.