CASE WESTERN RESERVE UNIVERSITY

STATISTICS COLLOQUIUM

Abstract

This talk focuses on aspects of several attempts to improve on ordinary kernel density estimation. Two general approaches have involved either higher-order kernels or locally adaptive smoothing. Current technology has not led to a clear distinction between these ideas. For example, some adaptive algorithms for positive kernels turn out to have a higher-order rate of convergence. In this talk, we describe new research in locally adaptive kernel smoothing that is not higher-order, and relate it to other work. The unusual behavior in the tail will be discussed in detail to provide an intuitive understanding.