Multi-dimensional Functional Principal Component Analysis

Abstract

Functional principal component analysis is one of the most commonly employed approaches in functional and longitudinal data analysis and we extend it to analyze functional/longitudinal data observed on a general $d$-dimensional domain. The computational issues emerging in the extension are fully addressed with our proposed solutions. The local linear smoothing technique is employed to perform estimation because of its capabilities of performing large-scale smoothing and of handling data with different sampling schemes (possibly on irregular domain) in addition to its nice theoretical properties. Besides taking the fast Fourier transform strategy in smoothing, the modern GPGPU (general-purpose computing on graphics processing units) architecture is applied to perform parallel computation to save computation time. To resolve the out-of-memory issue due to large-scale data, the random projection procedure is applied in the eigendecomposition step. We show that the proposed estimators can achieve the classical nonparametric rates for longitudinal data and the optimal convergence rates for functional data if the number of observations per sample is of the order $(n/\log n{)}^{d/4}$. Finally, the performance of our approach is demonstrated with simulation studies and the fine particulate matter (PM 2.5) data measured in Taiwan.

References in corpus (6)

• Uniform convergence rates for nonparametric regression and principal component analysis in functional/longitudinal data
• Methodology and theory for partial least squares applied to functional data
• Dynamical functional prediction and classification, with application to traffic flow prediction
• Evaluating stationarity via change-point alternatives with applications to fMRI data
• Detecting and handling outlying trajectories in irregularly sampled functional datasets
• A two-way regularization method for MEG source reconstruction

Cited by in corpus (4)

• Low-Rank Covariance Function Estimation for Multidimensional Functional Data
• Neural Networks for Scalar Input and Functional Output
• Spectrophotometric calibration of low-resolution spectra
• Principal Component Analysis of Two-dimensional Functional Data with Serial Correlation