Wednesday, October 21st, 2015
3:00pm - 4:00pm, in McCormack 2-205

David Degras-Valabregue

UMB/DePaul

Online Principal Component Analysis in High Dimension: Which Algorithm to Choose?

Abstract: In the current context of data explosion, online techniques that do not require storing all data in memory are indispensable to routinely perform tasks like principal component analysis (PCA). Recursive algorithms that update the PCA with each new observation have been studied in various fields of research and found wide applications in industrial monitoring, object tracking, astronomy, and latent semantic indexing, among others. In this talk, I will present the principal approaches to online PCA, namely, perturbation techniques, incremental methods, and stochastic optimization. A simulation study will compare the statistical accuracy, computational speed, and memory usage of related algorithms. I will also discuss the case of functional data and show applications of online PCA to data compression and face recognition using high-dimensional data. All considered algorithms are available in the package onlinePCA on CRAN.