What the training of a neuronal network optimizes

Area of Science:

• Computational neuroscience
• Complex systems modeling

Background:

• Investigating artificial neuronal networks is crucial for understanding brain function and developing AI.
• Watts-Strogatz networks offer a model for complex network topology with properties between regular and random networks.

Purpose of the Study:

• To investigate a training model for neuronal networks composed of integrate-and-fire neurons.
• To analyze how network training affects signal transmission, reaction time, and energy efficiency.

Main Methods:

• A computational model of neuronal networks with Watts-Strogatz topology was developed.
• The model utilized integrate-and-fire neurons and a 'kick-and-delay' rule for synaptic modification during training.
• External stimuli were applied to a single receptor neuron to initiate network activity.

Main Results:

• Network training significantly decreased the mean path length for signal transmission from the receptor neuron.
• A reduction in reaction time and energy expenditure for the network to respond to stimuli was observed.
• The initial distribution of synaptic couplings was identified as a critical factor influencing the performance of trained networks.

Conclusions:

• The 'kick-and-delay' training method effectively optimizes neuronal network performance.
• Network topology and initial synaptic configurations are key determinants of efficient information processing in trained networks.
• This model provides insights into adaptive network dynamics and efficient signal processing in biological and artificial systems.