FNS allows efficient event-driven spiking neural network simulations based on a neuron model supporting spike latency

Area of Science:

• Computational Neuroscience
• Neurotechnology
• Artificial Intelligence

Background:

• Spiking neural networks (SNNs) are crucial for brain modeling but require significant computational resources.
• Efficient neural models are needed for low-power applications like neuroprosthetics.

Purpose of the Study:

• Introduce FNS, a novel framework for efficient SNN simulations based on the Leaky Integrate-and-Fire with Latency (LIFL) model.
• Enable detailed neural simulations with reduced computational and memory footprints.

Main Methods:

• Developed FNS, an event-driven SNN framework incorporating LIFL neuron models.
• Implemented multi-thread parallelization and periodic dumping for precise, efficient simulations.
• Defined heterogeneous neuron groups, multi-scale connectivity, delayed connections, and plastic synapses.

Main Results:

• FNS demonstrated superior performance in simulation time and memory usage compared to NEST.
• The framework supports complex network structures and long-timescale simulations.
• Achieved efficient modeling of neuronal dynamics with reduced resource requirements.

Conclusions:

• FNS provides an efficient and powerful tool for SNN-based brain modeling, particularly for resource-constrained applications.
• The framework facilitates exploration of neural population dynamics.
• LIFL-based modeling offers a promising balance between biological realism and computational efficiency.