Exploring and understanding the dynamics on simple recurrent rate networks in the presence of noise
Keywords
Loading...
Authors
Issue Date
2022-07-07
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Artificial recurrent neural networks simulate the way biological systems operate.
The input is spread out over a set of interconnected artificial neurons,
and signals are spread through the network. These kinds of networks are
known to be capable of being trained to be very robust to the effects of
noise, but the internal mechanics of this noise remains poorly understood.
To improve the understanding of the effect of noise, I will study the effects
of noise by creating networks with 2D input and 2D output, adding noise to
these networks, and making geometric plots of the output of these networks.
Using these geometric plots I will explain what happens in the network and
what kind of effect the noise has. This will create an intuition of what kind
of effect noise can have on a network, and should be generalizable to larger,
more complex networks. This intuition can lead to future applications of
recurrent neural networks to be built with the effects of noise in mind, and
enhance the capability to minimize the effects of noise.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen