Adaptive Stimulus Selection in Estimating Vestibular Model Parameters in the Rod-and-Frame Task
The perception of upright can be biased by the visual context. The extent of the bias is linked to the functioning of the vestibular organ. The rod-in-frame task is an excellent method to test this biasing effect. The response to the task can be fitted by a 4-parameter Bayesian model. In this thesis project, an algorithm that adaptively selects stimuli based on entropy was implemented (in Python) to speed up the convergence of the model’s parameters. Tests with a generative agent showed convergence of some of the parameters after only 500 trials (in comparison with ±1600 trials). Experimental tests with real subjects show similar results. Although there are many aspects that could be improved, the results indicate that adaptive stimulus selection can indeed significantly reduce the number of trials needed for convergence of the model’s parameters. With further research, there might be a possibility to use the model with adaptive stimulus selection as a clinical tool to help patients with vestibular deficiencies.
Faculteit der Sociale Wetenschappen