Multi-alternative Adaptive Stimulus Selection Applied to Estimation of Visual–Vestibular Parameter Interactions in the Rod-and-Frame Task

No Thumbnail Available
Journal Title
Journal ISSN
Volume Title
Visual–vestibular interactions in the rod-and-frame task have shown that subjects flexibly weigh visual panoramic and vestibular information based on their orientation-dependent reliability. This flexible weighing of visual and vestibular information results in observed verticality biases and associated response variabilities. The response to the rod-and-frame task can be fitted by a psychometrical estimate of a Bayesian optimal integration model to the acquired or simulated data. In this thesis, an estimate of the parameters of a generative Bayesian multisensory integration model was implemented (in Python) which e.g. represents a simulated subject. For the implementation of this model, an adaptive stimulus selection model was used that selects (optimal) stimuli based on an entropy measure, to select multiple-trials-ahead into the future. Optimal stimuli selection is a computational trick to speed up the convergence and accuracy of the model parameters estimation. Simulation testing with the backward induction algorithm showed signs of convergence of some of the parameters after 250 trials by optimizing multiple-trials-ahead. These results show great promise for optimizing multiple trials ahead in stimulus selection processes to reduce the number of trials needed for convergence of the model’s parameters in the rod-and-frame task. New opportunities might open up with this next step taken in more efficiently determining the parameters of the visual–vestibular interactions model of the rod-and-frame task, for medical tools, which are needed for patients with deterioration in the vestibular organ which impairs their eyesight.
Faculteit der Sociale Wetenschappen