Common Principal Component Analysis (series unlisted)

Abstract: Dimensionality reduction attempts to transform often high dimensional data into a lower dimensional representation while maintaining the data's intrinsic properties. Several methods have been developed to accomplish this task, but perhaps the most widely used is Principal Component Analysis (PCA). While PCA is well known, its extension to multiple populations, Common Principle Component Analysis (CPCA), is much lesser known. In this talk we introduce CPCA and discuss its efficacy for completing dimensionality reduction across multiple populations. In addition, we discuss spectral approaches for fitting CPCA in practice, including randomized algorithms for truncated singular value decompositions. Finally, we employ CPCA for simultaneous dimensionality reduction across penguin species in the Palmer Penguin dataset.

Mathematics

Audience: researchers in the topic

BU Community Seminar