Eye movement data analysis using Gaussian Process
Code, data and material for my presentation in Bayes@Lund 2017.
Statistical Inferences of Eye movement data using Bayesian smoothing
Human observer performs rapid ballistic eye movements to sample visual information, with a combination of fixations and saccades. One of the challenges in the analysis of eye gaze is the sparseness of the data, as only one sample is observed at a given time point. One of the solutions is the usage of kernel smoothing. It is first applied as descriptive data representation (i.e., heat map), and later for statistical inference (e.g., iMap4, Lao et al., 2016). However, it is impossible to infer the smoothing parameters (e.g., kernel size), as they are mostly fixed and chosen arbitrarily. A more natural solution is to use Gaussian Process. Bayesian inference on the kernel length scale of the covariance function can capture the actual smooth spatial-temporal effect, whereas the uncertainty of the observed data points is naturally expressed as the kernel variance. Here, I demonstrate the advantage of this approach on an eye movement study using dynamic stimuli.