A coupling network for brain computing: E-I balanced embedding in dual-attractor dynamics systems.

Area of Science:

• Computational Neuroscience
• Neural Network Modeling
• Brain-Inspired Computing

Background:

• Continuous attractor neural networks (CANNs) are crucial for modeling continuous variable representation in the brain.
• Real brain networks exhibit complex, non-random neuronal connections and interactions.
• Understanding multi-neuron group interactions is key to deciphering complex brain network behavior.

Purpose of the Study:

• To propose a selective coupling network model based on CANN for studying complex brain networks.
• To investigate the interaction between two CANN classes under fast excitation-inhibition (E-I) balance for motion direction recognition.
• To explore the role of complex network interactions in brain information processing.

Main Methods:

• Developed a selective coupling network model integrating Continuous Attractor Neural Networks (CANNs).
• Simulated interactions between two CANNs with differing selective preferences under dynamic E-I balance.
• Employed theoretical analysis alongside simulation results to interpret network dynamics.

Main Results:

• Fast E-I balance facilitates indirect linking effects among multiple neural networks.
• Indirect mutual inhibition within coupled networks enhances response accuracy and stability for specific positions.
• Demonstrated that E-I balance can mediate collaborative activity between distinct neural networks.

Conclusions:

• Dynamic E-I balance in the brain enables indirect coupling and coordination between neural networks.
• This mechanism enhances the stability and accuracy of neural representations.
• Findings offer insights into multi-network coupling for developing brain-like computing systems.