Title: Modeling motor learning using heteroskedastic functional principal components analysis
Authors: Daniel Backenroth - Columbia Mailman School of Public Health (United States)
Michelle Harran - Columbia University (United States)
Juan Cortes - Columbia University (United States)
John Krakauer - Johns Hopkins University (United States)
Tomoko Kitago - Columbia University (United States)
Jeff Goldsmith - Columbia University (United States) [presenting]
Abstract: Experiments involving kinematic data -- dense recordings of hand or finger position over time during the execution of a motion -- provide deep insights into the processes underlying motor control and learning. To model the reduction of motion variance achieved through repetition, we extend the functional principal components analysis framework to allow subject and covariate effects on score variances. In a setting where the components are invariant across subjects and covariate values, this approach provides a flexible and interpretable way to explore factors that affect the variability of functional data. Parameters are jointly estimated in a Bayesian framework using both MCMC and a computationally efficient variational approximation.