FORCE training a Neurobiologically-Motivated Network for Speech Recognition
Keywords
Loading...
Authors
Issue Date
2022-08-20
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
How language is processed in the brain is currently not well understood.
With the use of neurobiologically-motivated models, brain functions like
these can be better studied. Supervised learning methods provide one way
to shape the network dynamics of such models. One such method is FORCE
training. Based on concepts from reservoir computing, FORCE has been
used to reproduce various complex behaviours. In this project, a spiking
neural network with realistic neural dynamics is FORCE-trained to classify
spike-encoded speech input. It was found that FORCE training does not
enhance word recognition in the network. Based on the idea that FORCE
training works best when replicating repetitious input, further tests were
done with a different input structure, as well as tests with FORCE-training
a High Dimensional Temporal Signal. The utility of FORCE training appeared
to improve when reproducing a repetitious input, although these
results are too limited to be definitive.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen