Many contemporary large-scale applications involve building and reproducing interpretable models linking a large set of potential covariates to some response of interest in a nonlinear fashion. Although this modelling problem has been studied extensively, it has remained largely unclear how to control the fraction of false discoveries effectively even in high-dimensional logistic regression, not to mention general high-dimensional nonlinear models. To address this practical problem, Fan and colleagues recently proposed a new framework of model-x knockoffs. In this talk, Dr. Fan will give a high-level overview of the model-x knockoffs framework. She will present a real data application and discuss a DNN architecture for integrating the model-x knockoffs framework with deep learning models.
Speaker: Yingying Fan, PhD, Professor of Economics and Data Science, USC
Register: http://eepurl.com/g1X35P