2317
Plays
28 Nov 2020
Abstract:
Unsupervised Variable Selection is an important step in many data analysis applications, either as an objective in its own right, for example, for sensor selection and data visualisation, or as a pre-processing step to address the curse of dimensionality and collinearity issues that arise with high dimensional data. Principal Component Analysis (PCA) is a well-established technique for dimensionality reduction which supports many of these applications, but it lacks transparency in that the reduced set of variables produced are linear combinations of the all the original variables, making the most significant variables difficult to determine. In this keynote I will introduce Forward Selection Component Analysis (FSCA), which was developed as a counterpart to PCA for unsupervised variable selection. FSCA seeks to identify a small number of key variables that are representative of the observed variance across all variables in a dataset. FSCA was initially introduced in the context of Optical Emission Spectroscopy data analysis of plasma etch processes where isolating a small number of wavelengths is important for understanding the underlying plasma chemistry. More recently, it has been found to be a particularly effective tool for optimising measurement site selection for spatial wafer metrology in semiconductor manufacturing. In addition to presenting these examples, I will briefly touch on recent extensions of the basic FSCA algorithm including the Forward Selection Independent Variable (FSIV) algorithm which is orientated towards variable selection for anomaly detection. I will also discuss how results from the field of combinatorial optimisation can be used to achieve efficient algorithm implementations.
京公网安备 11010502051620号
