Towards accessible Parallel Discrete Event Simulation of Spiking Neural Networks

Abstract

Spiking Neural Networks (SNNs) are a class of Artificial Neural Networks that closely mimic biological neural networks. Their potential to advance medical and artificial intelligence research makes them particularly interesting to study. Since their behaviour cannot be computed with single one-shot functions, simulations are employed to study their evolution over time. Recent works presented the possibility of simulating SNNs using speculative Parallel Discrete Event Simulation (PDES). However, no high-level interface to run SNN simulations using PDES was provided, leaving the model implementation to the users. This demanding process creates a barrier to the adoption of the method. In this work, the initial efforts towards making PDES-based simulation of SNNs easily accessible via interfaces with a high abstraction level (PyNN) are reported. Preliminary performance results are reported and comparisons are made between PDES using the ROme OpTimistic Simulator (ROOT-Sim), and the state-of-the-art SNN simulator NEST, both used through the PyNN interfaces.