References & Citations
Statistics > Methodology
Title: Bayesian hierarchical functional data analysis with basis function approximations using Gaussian-Wishart processes
(Submitted on 23 Dec 2015 (v1), revised 9 Jun 2016 (this version, v2), latest version 12 Dec 2016 (v3))
Abstract: Functional data are defined as realizations of random functions varying over a continuum, which are usually collected on discretized grids but not necessarily on common grids with low dimension. To tackle the issue of uncommon observation grids and high dimensionality for Bayesian functional data analysis with Gaussian processes, we propose conducting posterior inferences with basis function approximations. Specifically, we illustrate our basis function approximation method for a Bayesian hierarchical model that simultaneously smooth functional observations and estimate the mean-covariance functions using Gaussian-Wishart processes. Assuming functional data are of Gaussian processes and approximated by a system of basis functions, we can conduct Bayesian inferences for the functional signals, mean, and covariance through the posterior samplings of the basis function coefficients. Both simulations and real studies demonstrate that, our approach with basis function approximations could conduct posterior inferences $>10$ times more efficient on both computation speed and memory usage, providing stable and consistent posterior estimates. When the functional data have low dimensional common grids, our approach generates similar results as the usual Bayesian inferences. When the observation grids are high dimensional or random, our approach can still efficiently conduct inferences while the usual Bayesian inferences suffer serious computation burden and unstableness. Our basis function approximation method sheds lights on how to resolve the curse of dimensionality for Bayesian functional data analysis.
Submission history
From: Jingjing Yang [view email][v1] Wed, 23 Dec 2015 18:09:34 GMT (1086kb,D)
[v2] Thu, 9 Jun 2016 17:01:49 GMT (1762kb,D)
[v3] Mon, 12 Dec 2016 18:52:03 GMT (1763kb,D)
Link back to: arXiv, form interface, contact.