Accelerating spiking neural network simulations with PymoNNto and PymoNNtorch.

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Neuromorphic Engineering

Background:

• Spiking neural network (SNN) simulations are crucial for advancing research in computational neuroscience, AI, and neuromorphic engineering.
• Existing SNN simulators offer diverse functionalities but can have limitations in custom code integration and GPU acceleration.
• PymoNNto is a Python-based toolbox known for its modularity and flexibility in embedding custom code for SNN simulations.

Purpose of the Study:

• Introduce PymoNNtorch, a PyTorch-native framework for SNN simulations.
• Retain the modular and flexible design of PymoNNto.
• Leverage native GPU support and optimized operations for enhanced simulation performance.

Main Methods:

• Developed PymoNNtorch with a native PyTorch backend, building upon PymoNNto's modular architecture.
• Implemented common network operations using PyTorch for efficient GPU utilization.
• Compared PymoNNtorch's performance against established simulators like NEST, ANNarchy, and Brian 2.

Main Results:

• PymoNNtorch demonstrates significant speed-ups in SNN simulations due to its native PyTorch implementation and GPU acceleration.
• The modular design inherited from PymoNNto is maintained, facilitating ease of use and custom code integration.
• Optimized indexing operations further contribute to the enhanced performance of PymoNNtorch.

Conclusions:

• PymoNNtorch offers a powerful and efficient tool for SNN research by combining modularity with native GPU acceleration.
• The framework facilitates faster development and execution of complex SNN models in the Python programming language.
• PymoNNtorch represents a valuable advancement for researchers in computational neuroscience, AI, and neuromorphic engineering seeking high-performance simulation tools.