Using Markov Chain Monte Carlo for Parameter Estimation in a Visual-Vestibular Integration Model

Keywords

Loading...
Thumbnail Image

Issue Date

2023-07-01

Language

en

Document type

Journal Title

Journal ISSN

Volume Title

Publisher

Title

ISSN

Volume

Issue

Startpage

Endpage

DOI

Abstract

There has been research in the past that developed models of how the brain can use multisensory information to estimate the perception of the vertical. Alberts et al. (2016) proposed a Bayesian optimal integration model for verticality perception in the rod-and-frame task. This model finds uncertainty estimates for its parameters using the maximum a posteriori (MAP) probability estimates optimized using gradient descent and bootstrap sampling methods. The goal of this thesis is to improve the parameter estimation procedure by implementing a Markov Chain Monte Carlo (MCMC) sampling approach for a modified version of the model from Alberts et al. (2016). It is hypothesized that the MCMC method is an improvement, providing more consistent parameter estimates and greater resilience to optimization problems. Generative data is used to evaluate parameter recovery. The results show that the MCMC sampler provides more precise parameter estimates across chains and iterations, where the distributions on the prior yield stability. The MCMC approach also revealed a reliable uncertainty quantification through the generated full posterior distributions. In contrast, the gradient descent-based bootstrap MAP method suffers from local optima traps and, for specific parameters, high uncertainty. In summary, this thesis establishes a foundation for building more efficient Bayesian optimal integration models with MCMC. The proposed method could facilitate clinical applications of visual-vestibular integration models for verticality perception to describe multisensory functioning.

Description

Citation

Faculty

Faculteit der Sociale Wetenschappen