Non-trivial relationship between behavioral avalanches and internal neuronal dynamics in a recurrent neural network

Area of Science:

• Computational Neuroscience
• Complex Systems
• Dynamical Systems

Background:

• Neuronal activity and behavior are interconnected in a closed-loop system.
• Previous studies suggested a direct relationship between scale-free neuronal and behavioral dynamics.
• These studies often focused on local network dynamics, limiting a comprehensive understanding.

Purpose of the Study:

• To investigate the relationship between internal neuronal dynamics and output statistics in a model system.
• To determine if power-law distributions in behavior correlate with specific neuronal dynamics.
• To explore the robustness of neuronal dynamics under perturbations.

Main Methods:

• Utilized a recurrent neural network (RNN) initialized in a chaotic state.
• Trained the RNN to produce behavioral states with power-law duration distributions.
• Analyzed neuronal avalanche size and duration distributions in trained and randomized network configurations.

Main Results:

• Training altered internal network dynamics, producing power-law neuronal avalanche size distributions.
• Randomizing network connectivity changes largely preserved power-law features in neuronal avalanche size distributions.
• Neuronal avalanche size distributions were invariant, even when behavioral power-law dynamics were not preserved.

Conclusions:

• A direct one-to-one correspondence between behavioral and neuronal statistical features is not established.
• The relationship between neuronal dynamics and behavior is non-trivial and complex.
• Intrinsic neuronal dynamics' statistical properties can be preserved despite perturbations in output-generating mechanisms.