PyGeNN

PyGeNN accelerates simulation of spiking neural networks (SNNs) on GPUs by providing a Python interface to the GeNN C++ library that generates efficient GPU-targeted simulation code.

Key Features:

• GPU Acceleration: Uses GPUs to execute SNN simulations, providing substantial speed-ups compared to CPU-based execution.
• Python Integration: Exposes GeNN functionality through Python while generating C++ simulation code for GPU execution.
• Efficient Spike Recording: Implements an optimized spike recording system that can reduce runtime overheads by up to 10 times compared to previous methods.
• Performance Benchmarks: Enables simulation of complex neural models faster than real-time neuromorphic systems, including full-scale cortical column models.
• Scalability and Flexibility: Supports long-duration simulations, intricate stimuli, and custom learning rules, and can simulate models with complex three-factor learning rules almost two orders of magnitude faster than real-time.

Scientific Applications:

• Computational Neuroscience: Large-scale SNN simulations for investigating network dynamics and spike-based computation.
• Cortical Column Modeling: Simulation and analysis of full-scale cortical column dynamics.
• Learning Rule Development: Implementation and evaluation of custom and complex learning rules, including three-factor learning rules.
• Spike-Data Experiments: High-fidelity spike recording and analysis for experiments that require detailed spike output.

Methodology:

PyGeNN interfaces with the GeNN C++ library to generate GPU-targeted SNN simulation code and implements an optimized spike recording system.