Feasibility Study of Online Learning in Spiking Neural Networks compared to Artificial Neural Networks
Keywords
Loading...
Authors
Issue Date
2022-06-19
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Real-world applications such as the Internet of Things provide a need for
the real-time, incremental, low-power online learning of high velocity data
streams. Whereas artificial neural networks (ANNs) are a popular choice
for machine learning problems, the more biologically plausible spiking neural
networks (SNNs) have properties that are interesting for this online learning.
In this review, several online ANN and SNN methods are compared
to address whether online learning is feasible in SNNs rather than ANNs.
It is found that both ANNs and SNNs are able to learn incrementally from
data streams, adapt to changes in the distributions of these data streams
(concept drift), and dynamically optimize their network structure. Whereas
ANNs have more efficient methods for training deep networks, the spikebased
characteristics of SNNs allow for sparse, event-driven and scalable
implementation on neuromorphic hardware, in turn leading to fast and energy
efficient online learning. This makes SNNs uniquely suitable for realtime
online learning when compared to ANNs.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen