Efficient Deployment of Spiking Neural Networks on SpiNNaker Neuromorphic Platform

Spiking Neural Networks (SNNs) have emerged as serious competitors of the traditional Convolutional Neural Networks (CNNs), as they unlock new potential of implementing less complex and more energy efficient neural networks. Current deep CNNs can be converted to SNNs for fast deployment on neuromorphic devices, however existing methods do not investigate the impact of hardware-related parameters that directly affect the accuracy of an SNN. In this brief, we target the SpiNNaker neuromorphic platform and we demonstrate a fast exploration framework that effectively decides the configuration of the target board, in order to achieve the highest possible accuracy. Experimental results show that our method reaches 98.85% SNN accuracy on MNIST dataset, while reducing the exploration time by a factor of 3× compared to exhaustive search.

I. Introduction

With the recent advancements in the area of machine learning, Spiking Neural Networks (SNNs) have gained a lot of attention as they promise better performance and lower energy consumption compared to traditional Convolutional Neural Networks (CNNs). SNNs resemble biological neurons as they utilize discrete spikes to decide whether a neuron is activated, and then propagate the information to other neighbor neurons through their synapses [1]. One approach to implement an SNN is to convert an existing CNN to SNN by mapping the CNN learning parameters to neurons, populations, and synapses. This approach is advantageous as it utilizes well established techniques and frameworks for fast CNN training on GPUs, and has shown promising results [2]. However, prior research works have not investigated the impact of hardware-related parameters on SNN accuracy when deploying a converted SNN on a neuromoprhic platform.