Three tools for the real-time simulation of embodied spiking neural networks using GPUs

Area of Science:

• Computational Neuroscience
• Robotics
• Artificial Intelligence

Background:

• Biologically-inspired spiking neural networks (SNNs) offer a powerful paradigm for understanding brain function and developing advanced AI.
• Simulating large-scale SNNs with millions of connections in real-time presents significant computational challenges.
• Integrating SNN simulations with real-world applications, such as robotics, requires specialized tools for data conversion and visualization.

Purpose of the Study:

• To present a comprehensive toolbox for constructing, simulating, visualizing, and deploying large-scale, biologically-inspired SNNs.
• To enable real-time simulation and 3D visualization of SNN activity.
• To facilitate the connection of SNNs to robotic systems and other external devices.

Main Methods:

• NeMo: A high-performance simulator for parallel SNN simulations on GPUs or multi-core processors, supporting various neural and oscillator models.
• SpikeStream: A visualization and analysis environment for constructing, storing, and visualizing SNNs and their activity in 3D.
• iSpike: A library for biologically-inspired conversion between real-world data and spike representations, crucial for robotic integration.

Main Results:

• The presented toolbox allows for the construction of SNNs with tens of thousands of neurons and millions of connections.
• Simulations can be performed in real-time, with activity visualized in 3D.
• The tools facilitate seamless integration with robotic platforms like the iCub.

Conclusions:

• The NeMo, SpikeStream, and iSpike tools provide a cohesive and powerful solution for advanced SNN research.
• These tools empower researchers to build, simulate, and deploy complex SNNs for diverse applications.
• The independent usability and synergistic integration of these tools enhance their utility in neuroscience and robotics.