mlGeNN: accelerating SNN inference using GPU-enabled neural networks
mlGeNN: accelerating SNN inference using GPU-enabled neural networks
复制标题
DOI：
10.1088/2634-4386/ac5ac5
复制
发表时间：
2022-03
期刊：
影响因子：
--
通讯作者：
J. Turner;James C. Knight;Ajay Subramanian;Thomas Nowotny
中科院分区：
文献类型：
--
作者：
J. Turner;James C. Knight;Ajay Subramanian;Thomas Nowotny
在本文中，我们提出了mlGeNN-一个Python库，用于将Keras中指定的人工神经网络（ANN）转换为尖峰神经网络（SNN）。SNN使用GeNN进行模拟，扩展以有效地支持卷积连接和卷积。我们在CIFAR-10和ImageNet分类任务上评估了转换后的SNN，并将其性能与原始ANN和其他SNN模拟器进行了比较。我们发现，使用在CIFAR-10数据集上训练的VGG-16模型执行推理比BindsNet快2.5倍，当使用在CIFAR-10上训练的ResNet-20模型时，使用FewSpike ANN到SNN转换，mlGeNN仅比TensorFlow慢2倍多一点。
In this paper we present mlGeNN—a Python library for the conversion of artificial neural networks (ANNs) specified in Keras to spiking neural networks (SNNs). SNNs are simulated using GeNN with extensions to efficiently support convolutional connectivity and batching. We evaluate converted SNNs on CIFAR-10 and ImageNet classification tasks and compare the performance to both the original ANNs and other SNN simulators. We find that performing inference using a VGG-16 model, trained on the CIFAR-10 dataset, is 2.5× faster than BindsNet and, when using a ResNet-20 model trained on CIFAR-10 with FewSpike ANN to SNN conversion, mlGeNN is only a little over 2× slower than TensorFlow.