Tutorial/Seminar: "An introduction to Principal Component Analysis (PCA) for high dimensional data, plus topics in PCA consistency and fast bootstrap computations"
Aaron Fisher, PhD Candidate
PhD candidate in Biostatistics, Johns Hopkins Bloomberg School of Public Health
An introduction to Principal Component Analysis (PCA) for high dimensional data, plus topics in PCA consistency and fast bootstrap computations
Abstract:
This casual presentation will include an introduction to principal component analysis (PCA) as a method for summarizing high dimensional (HD) data (e.g. brain images or genomic data). I will also survey a few results on conditions under which sample principal components (PCs) can have poor performance -- specifically when they can diverge from the population PCs as dimension increases. Finally, I'll talk some about my own research on PCA, which focuses on fast computational methods for estimating standard errors of sample PCs. These methods are based around a bootstrap procedure, and can reduce computation time from days to minutes compared to standard bootstrap methods.
Bio:
Ìý