Effects of Level of Detail on Predictive Processing
Keywords
No Thumbnail Available
Authors
Issue Date
2019-01-28
Language
en
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
According to the predictive processing account, the human brain tries to predict
the world around it, and revises its own prediction based on the prediction errors
these predictions result in. Immature models with no or very little experience to
base their predictions on, will inevitable make inaccurate predictions resulting in
high prediction errors. Reducing the prediction error of the predictive models is
always important, but arguably even more so in the case of these immature
models. One way of reducing the prediction error of a model is by reducing the
level of predictive detail, meaning that the resolution of what it predicts is
lowered. Because this reduces the amount of uncertainty about what it predicts,
the prediction error should drop accordingly. This is of course not without its
downsides, because a lower level of predictive detail will mean the amount of
information that be gained from the model is reduced along with the prediction
error. This thesis tried to answer two questions: “What is the difference in the size
of the early prediction error between predicting at a high level of detail (control
scenario) and predicting at a low level of detail (test scenario), and will switching
to a higher level of detail while learning result in a lower overall prediction error?”
As expected, the decrease in the prediction error of the test scenario predicting
at a lower level of detail was found to be significant. The effect on the later
prediction error was not found to be significant. However, the environment used
in this thesis was far from complex, and more experiments in such complex
environments would thus be desired.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen