【Academic lecture】Envelope-based partial least squares

Topic: Envelope-based partial least squares

Speaker:

Zhihua  Su is Associate Professor in Department of Statistics at the University  of Florida. She got her PhD in 2012 from University of Minnesota, Twin  Cities. Her research interest includes Feature  selection, Dimension reduction, Multivariate analysis, Bayesian  statistics, Algorithm development, Software development, Applied  statistics.

Introduction: 

Partial  least squares (PLS) is widely used in applied sciences an alternative  method to ordinary least squares (OLS) for estimating the regression  coefficients.  It is known that PLS often has a better prediction  performance compared to OLS, and the PLS algorithms can be adapted  directly to the n < p case.  Despite its popularity, the theoretical  properties of the PLS estimator are largely unknown.  As a result, it is  hard to determine when PLS is better than OLS, what are the limitations  for PLS and how to improve PLS.  Cook et al. (2013) built a connection  between PLS with a dimension reduction method called the envelope model.  They showed that at the population level, PLS and the envelope model  have the same target parameter, but they use different algorithms for  estimation. This connection allows PLS to be studied in a traditional  likelihood framework and facilitates model developments.  We will  address three issues of PLS in this context: variable selection,  categorical predictors and scale invariance.

Time: 2024.6.22 14:00