A1012
Title: Reproducible learning for censored data via deep knockoffs
Authors: Daoji Li - California State University Fullerton (United States) [presenting]
Abstract: A new feature selection procedure with guaranteed false discovery rate (FDR) control for censored data is introduced. By using deep knockoffs, the proposed procedure can handle covariates with arbitrary and unspecified data distributions. It also can deal with both continuous and categorical covariates. We provide theoretical justifications by showing that the FDR is controlled at the target level. Extensive numerical experiments confirm the generality, effectiveness, and power of our method.
