Parameter Recovery Analysis of the Bayesian Optimal Integration Model for Visual-Vestibular Interactions
Keywords
No Thumbnail Available
Authors
Issue Date
2019-07-04
Language
en
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The otolith organs in the vestibular system allows humans perceive linear acceleration as well as the
direction of gravity. Due to this property they are involved in processes like image stabilization during
motion and determining the head-in-space position relative to gravity. However, with age the performance
of the vestibular system decreases as the noise level of the vestibular signal in those processes
increases. Such a de cit results in a re-weighting of the visual and vestibular modalities involved. That
causes elderly to rely more on visual contextual cues. That is why an e cient and reliable test is needed
to measure the de cit. The psychophysical Bayesian optimal integration model for visual-vestibular
interactions by Alberts et al. (2016) can be used to quantify the sensory re-weighting. The rod-andframe
paradigm can generate the necessary data to t the model. The aim of this thesis is to determine
how well the recovered parameters of the tted model actually re
ect the true parameters. So far only
responses of human subjects have been used to t the model. Since the true parameters of humans
are unknown, it was not possible to validate the recovered parameters. By using a generative model in
silico a parameter recovery analysis is made possible. The parameters are recovered using a constrained
minimization of the negative log-likelihood to nd the optimal parameters given the data provided by
the generative model. The results show that the seven parameters of the model are recoverable with
high accuracy. For all parameters their true generative value is either equal to the mean of the recovered
values or was contained well within one standard deviation. Furthermore, the model tted with the
recovered parameters can account for the data of the generative model. Now that we have established
that parameters are in principle recoverable, future research should focus on how to reduce the number
of trials needed to recover the parameters. A promising direction for this is the use of adaptive stimulus
selection, as opposed to the pre-de ned, and probably sub-optimal, stimuli used in the current study.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen