Scholarlink

Although artificial neural networks (ANNs) are immensely powerful and have demonstrated their capabilities across many different applications and use cases [6, 9], their high power consumption and computational requirements often make them difficult to adopt for edge devices. In recent years, spiking neural networks (SNNs) have been proposed as an alternative with significantly lower power requirements. While there are already existing implementations of SNNs [2, 4, 5], most of these have been designed to handle only a moderate number of complex neurons. In this paper we will investigate the capabilities of feedforward networks based on a recently proposed hardware-efficient SNN architecture, showing that it can be used to classify images from the MNIST dataset with accuracies above 98%.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Influence of Spike Encoding, Neuron Models and Quantization on SNN Performance

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Daniel Windhager,
  • Bernhard Moser,
  • Michael Lunglmayr

摘要

Although artificial neural networks (ANNs) are immensely powerful and have demonstrated their capabilities across many different applications and use cases [6, 9], their high power consumption and computational requirements often make them difficult to adopt for edge devices. In recent years, spiking neural networks (SNNs) have been proposed as an alternative with significantly lower power requirements. While there are already existing implementations of SNNs [2, 4, 5], most of these have been designed to handle only a moderate number of complex neurons. In this paper we will investigate the capabilities of feedforward networks based on a recently proposed hardware-efficient SNN architecture, showing that it can be used to classify images from the MNIST dataset with accuracies above 98%.