Hybrid Neural Network Models Explain Cortical Neuronal Activity During Volitional Movement.

Area of Science:

• Computational neuroscience
• Neuroscience
• Artificial intelligence

Background:

• Large-scale neural networks exhibit massive interconnectivity, crucial for complex functions.
• Understanding neural connectivity is vital for both biological and artificial systems.
• Previous models struggled to capture dynamic statistical structures in neural networks.

Purpose of the Study:

• To develop hybrid neural network (HNN) models for analyzing neural connectivity.
• To investigate the statistical structure of connectivity in large-scale neural networks.
• To resolve questions regarding prescribed versus ongoing control of volitional movement.

Main Methods:

• Developed hybrid neural networks (HNNs) comprising artificial neurons.
• Trained a subset of artificial neurons to replicate experimental single-neuron responses.
• Utilized experimental firing rate data from monkey motor cortex during a reaching task.
• Analyzed dynamic statistics of neuron-neuron connections within trained HNNs.

Main Results:

• Trained recurrent and spiking HNNs exhibited state transitions mirroring empirical data.
• Identified that extrinsic input dynamics alter network connectivity to induce state transitions.
• Discovered two synaptic processes: widespread membrane potential buildup and specific action potential triggering.

Conclusions:

• HNNs effectively model realistic neuron-neuron connectivity and large-scale network functionality.
• The study provides foundational descriptions of how network dynamics drive volitional movement.
• Findings offer insights into the interplay between extrinsic inputs, synaptic dynamics, and network state changes.