CMStatistics 2018: Start Registration
View Submission - CMStatistics
B0953
Title: Median-of-means-type estimators and rates of convergence in normal approximation Authors:  Stanislav Minsker - University of Southern California (United States) [presenting]
Abstract: New results are presented for the class of estimators obtained via the generalized median-of-means (MOM) technique. These results stem from connections between performance of MOM-type estimators and the rates of convergence in normal approximation. We provide tight non-asymptotic deviations guarantees in the form of exponential concentration inequalities, as well as asymptotic results in the form of limit theorems. Our techniques will be illustrated with several examples, including the now-classical median-of-means estimator, and robust maximum likelihood estimation.