Watch a 41-minute lecture from the Simons Institute where Johannes Schmidt-Hieber from the University of Twente explores transfer learning of nonparametric least squares estimators under covariate shift. Dive into the convergence properties of empirical risk minimizers and discover how pointwise convergence rates are essential for deriving bounds under covariate shift. Learn how the nonparametric least squares estimator over 1-Lipschitz functions achieves minimax rate optimality with respect to a weighted uniform norm, demonstrating how the estimator adapts locally to design density despite being a global criterion. Examine specific convergence rates for various source/target density pairs and understand how weighting naturally accounts for non-uniform design distribution in domain adaptation applications.
Overview
Syllabus
Transfer learning via local convergence rates of the nonparametric least squares estimator
Taught by
Simons Institute