Towards Neuromorphic Computing on Edge: A Survey on Efficient Techniques for Spiking Neural Networks

Abdalla Rashed, Jorge Dias

Keywords: Spiking Neural Networks (SNNs), Neuromorphic Computing, Neuronal Dynamics in SNNs, ANN-SNN Conversion, Energy-Efficient Spiking Neural Networks

TL;DR: Spiking Neural Network techniques, based on neuronal and spiking dynamics, are surveyed and analyzed against ImageNet and CIFAR benchmarks for latency and energy-efficiency.

Abstract: Recent advancements in Spiking Neural Networks (SNNs) for machine learning have established them as an energy-efficient alternative to conventional Artificial Neural Networks (ANNs). However, training and deploying a deep SNN comes with complications related to propagation of various errors including vanishing gradients, ANN-to-SNN activation mismatch, and loss of transmitted information in the spiking patterns. This work surveys techniques to mitigate those errors and to increase the expressive capacity of the networks on the ground of spiking and neuronal dynamics. Latency vs accuracy results for the reviewed methods are reported on standard benchmarks such as ImageNet and CIFAR. Additionally, the temporal metric $T$ and its relationship to model efficiency is discussed.

Submission Type: Novel research

Student Paper: Yes

Demo Or Video: No

Public Extended Abstract: Yes

Submission Number: 1