Modelling BCI Learning
Keywords
Loading...
Authors
Issue Date
2017-08-24
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
The aim of this thesis is to investigate if it is possible to computationally model
BCI learning and use reinforcement learning to predict the effects of different neurofeedback
parameters. A model of BCI learning was constructed and experiments
were run with Q-learning agents using different model parameters. According to the
model, continuous feedback leads to signifi cantly more efficient BCI learning than
discrete feedback, and BCI learning becomes quadratically slower as the signal-to-noise
ratio decreases. Optimistic feedback results in signi ficantly faster learning than
neutral feedback, and pessimistic feedback signifi cantly slows down learning relative
to neutral feedback. Since research in this area is rather scarce, it is hard to compare
the results gained using the model with empirical data. However, the results do
seem to reflect existing data. The optimistic feedback data reflecting empirical data
seems to be the result of a side-effect of the way rewards are calculated. Overall,
the results suggest that the model may be used to predict the effects of different
neurofeedback parameters, but further research would be needed.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen