Analysis of the effects of decay coefficient and time resolution in SNN backpropagation
- Authors
- Um, Kwiseob; K.S.; Hwang, Heejae; H.; Kim, Hyungtak; H.; Heo, Seoweon; S.W.
- Issue Date
- 2020
- Publisher
- Institute of Electrical and Electronics Engineers Inc.
- Keywords
- Backpropagation; LIF neuron; Spiking neural network
- Citation
- 2020 International Conference on Electronics, Information, and Communication, ICEIC 2020
- Journal Title
- 2020 International Conference on Electronics, Information, and Communication, ICEIC 2020
- URI
- https://scholarworks.bwise.kr/hongik/handle/2020.sw.hongik/12468
- DOI
- 10.1109/ICEIC49074.2020.9051056
- ISSN
- 0000-0000
- Abstract
- High-performance neural networks operating at low power use spiking neural network (SNN) that are biologically closer than traditional ANN. SNN, unlike ANN, receives a series of binary-coded spike trains as input and updates the membrane potential of the neuron and generates spikes over a period of time specified by the number of spike trains. The function that generates the spike corresponding to the activation function of the ANN is not differentiable, which makes it difficult to apply the backpropagation (BP) algorithm used in the ANN. In order to overcome this problem, studies using numerical approximation of derivatives have been carried out in various ways. However, research on the decay coefficient and the number of spike trains, which are characteristic of SNN neuron, are insufficient. In this paper, we analyze the distribution of spikes and discuss how the decay coefficient characteristics of neurons and the number of spike trains affect network performance. © 2020 IEEE.
- Files in This Item
- There are no files associated with this item.
- Appears in
Collections - College of Engineering > School of Electronic & Electrical Engineering > 1. Journal Articles
Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.