Recurrent Feedback in a Neuron’s Dendrites
Keywords
Loading...
Authors
Issue Date
2022-06-17
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Artificial Neural Networks (ANNs) have applicabilities in many different
fields, such as facial recognition and predicting cancer. ANNs consist of
many simple computational neurons, which receive a set of inputs and compute
a single output. Biological neurons are, however, much more complex.
Biological neurons reside in a dynamic environment, learn non-locally and
have complex dendritic trees which receive various inputs and filter these in
different ways. The connections between the dendritic trees and the soma of
the neuron are bi-directional: somatic activity influences the dendrites and
vice-versa. All of these properties could be essential to the functioning of
biological neurons. In this study, a new, dynamic model of a single neuron
is presented, where the connections between dendrites and the soma are bidirectional.
For this research, the Bienenstock, Cooper and Munro (BCM)
and Oja’s rule are used as learning rules. The feedforward dynamic model
of a single neuron presented in this research shows a significant difference in
behaviour compared to a discrete ANN when the BCM rule is used. With
Oja’s rule, the behaviour of the presented feedforward dynamic model is
similar to the behaviour of discrete ANNs which use this rule. When recurrence
is introduced to the model, the systems become more stable and,
for the BCM rule, dominance of one or more dendrite neurons occurs. This
research provides a more biologically compatible model of single neurons,
which can be used to gain a better understanding of the complex functionalities
of biological neurons. Reproducing these essential functionalities in
a simplified model could improve ANNs by improving the functionality of
single neurons.
Description
Citation
Faculty
Faculteit der Sociale Wetenschappen