Event-driven simulations of a plastic, spiking neural network

Area of Science:

• Computational Neuroscience
• Artificial Neural Networks
• Complex Systems

Background:

• Leaky integrate-and-fire neurons are fundamental models in computational neuroscience.
• Spike-timing-dependent plasticity (STDP) is a key mechanism for synaptic modification in biological and artificial neural networks.
• Understanding network dynamics under plasticity is crucial for brain function and AI development.

Purpose of the Study:

• To investigate the impact of a plasticity parameter on network activity in a fully connected neural network.
• To analyze synaptic weight distributions across different plasticity regimes.
• To identify and characterize emergent network structures.

Main Methods:

• Event-driven simulations of finite-size networks (up to 128 neurons).
• Analysis of stationary synaptic weight conformations.
• Pseudophysical visualization techniques to identify network structures.

Main Results:

• Low plasticity parameter values result in noise-dominated activity.
• High plasticity parameter values lead to self-sustaining network activity.
• Synaptic weights distribute narrowly around the plasticity parameter in low/high regimes, broadening in the transition region.
• Emergent network structures of 'path' or 'hub' types were identified in the transition region.

Conclusions:

• The plasticity parameter significantly influences neural network dynamics, transitioning from noise to self-sustaining activity.
• Emergent network structures arise in the transition region, characterized by specific synaptic weight distributions.
• Findings align with mean-field theory predictions for synaptic weight distributions.