报告人：Prof. Linda Zhao

Statistics Department, University of Pennsylvania

Guanghua School, Beijing University

报告时间：2005年12月 28日下昼4:30

报告所在：伟易博楼120室

Abstract

This talk describes some nonparametric density algorithms using the root-unroot paradigm. The paradigm involves several easily implemented steps, as follows: Suitably bin the data. Calculate the square Root of the normalized, binned data. Apply a nonparametric regression estimator. Then “Unroot” in a suitable fashion. (Often “unroot” = square, especially if the original “root” step involves suitable minor adjustments.)

Asymptotic results will be given to show that the RU procedure involves only an insignificant loss of information.

Adaptive procedures are feasible, as are multivariate versions. Confidence bands can also be produced for the estimated density functions.

The methodology will be illustrated with data from a telephone call-center, as well as from Monte-Carlo experiments.