SpykeTorch: Efficient Simulation of Convolutional Spiking Neural Networks With at Most One Spike per Neuron.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Machine Learning

Background:

• Spiking neural networks (SNNs) offer hardware-friendly and energy-efficient alternatives for artificial intelligence (AI).
• Existing SNN simulation frameworks often lack the practical efficiency required for large-scale AI applications.

Purpose of the Study:

• Introduce SpykeTorch, an open-source, high-speed simulation framework for deep convolutional SNNs.
• Provide a PyTorch-based platform for efficient SNN simulations.

Main Methods:

• Developed SpykeTorch using PyTorch for tensor-based computations, enabling just-in-time optimization.
• Implemented simulation of convolutional SNNs with single spikes per neuron and rank-order encoding.
• Integrated learning rules including spike-timing-dependent plasticity (STDP) and reward-modulated STDP (R-STDP).

Main Results:

• SpykeTorch demonstrates high-speed simulation capabilities for SNNs.
• The framework is generic and can reproduce results from various studies.
• Optimized computations run efficiently on CPUs, GPUs, and multi-GPU platforms.

Conclusions:

• SpykeTorch addresses the efficiency limitations of current SNN frameworks for large-scale AI.
• The framework facilitates research and application of deep convolutional SNNs.
• SpykeTorch offers a flexible and performant tool for AI research.