Convex Loss Versus Semiparametric Likelihood Machine Learning Estimates

• Activity Number: 103 - Semiparametric Inference with High-Dimensional and Complex Data
• Type: Invited
• Date/Time: Monday, August 3, 2020 : 1:00 PM to 2:50 PM
• Sponsor: ENAR
• Author(s): Michael R. Kosorok*
• Companies: University of North Carolina at Chapel Hill
• Keywords: machine learning; convex loss; semiparametirc ; maximum likelihood; precision medicine

Abstract

For certain machine learning tasks, both semiparametric likelihood (SL) based estimators and convex loss (CL) based estimators are available. In some settings, the SL estimators are harder to compute but more accurate whereas the CL estimators are easier to compute but less accurate. In this presentation, we discuss this tension in some detail for a precision medicine application and provide some theoretical and simulation results which provide insight into this challenge.