Parameter Recovery of a Visual-Vestibular Interactions Model using Adaptive Stimulus Selection
The signal quality of the vestibular organ diminishes with age. This can have negative e ects on balance, coordination and vision. The aim of the project was to develop techniques that radically decrease the number of trials needed in the Rod-and-Frame task for testing vestibular functionality based on a Bayesian model. The model is a reduced version of the model by Alberts et al. (2016) where the head orientation is xed at upright. The mean inferred parameter values in that paper were used to make a generative model that generates responses given stimuli. Adaptive stimulus selection was used to select the next stimulus, such that the corresponding response was maximally informative for the model parameter recovery. This was done in silico, by adaptively manipulating the rod and frame orientations. The proposed model's parameters could not be successfully recovered. However, the advantage of adaptive stimulus selection over random stimulus selection was e ectively shown when recovering single parameter and parameter pair values. This advantage was not observed for the model with all free parameters, suggesting that the proposed model's parameters are fundamentally unrecoverable. The discussion provides a number of suggestions on how this unrecoverability might be overcome.
Faculteit der Sociale Wetenschappen