Event-driven simulation scheme for spiking neural networks using lookup tables to characterize neuronal dynamics

Area of Science:

• Computational Neuroscience
• Computational Neuroscience Modeling
• Neural Network Simulation

Background:

• Neuronal communication relies on action potentials (spikes), influencing short-term and long-term synaptic dynamics.
• Sparse action potential activity in the brain has led to event-driven simulation schemes to reduce computational cost.
• Existing event-driven schemes often use overly simplified neuronal models, limiting realism.

Purpose of the Study:

• To implement and critically evaluate an event-driven algorithm (ED-LUT) using precalculated look-up tables for neuronal and synaptic dynamics.
• To enable high-speed simulations with more complex and realistic neuronal models.
• To facilitate the study of synaptic plasticity and learning in large-scale neural networks.

Main Methods:

• Developed an event-driven algorithm (ED-LUT) utilizing precalculated look-up tables.
• Implemented exponential synaptic conductances to model shunting inhibition.
• Introduced an improved two-stage event-queue algorithm for efficient scaling in highly connected networks with propagation delays.

Main Results:

• The ED-LUT algorithm successfully characterized synaptic and neuronal dynamics, allowing for more complex models.
• Demonstrated the implementation of shunting inhibition, crucial for cellular computation.
• The improved event-queue algorithm efficiently scaled simulations for complex network structures.

Conclusions:

• The ED-LUT approach offers a significant advancement for high-speed, realistic neural network simulations.
• The method supports the inclusion of synaptic plasticity, paving the way for studying learning and adaptation.
• This work enhances the computational feasibility of exploring complex brain dynamics and functions.