Fred J. Hickernell (Illinois Institute of Technology)
Title: The Trio Identity for Monte Carlo Methods
Abstract: When the mean of a random variable cannot be computed analytically, it may be approximated by the sample mean. This is the essence of Monte Carlo methods. The error of this approximation can be decomposed exactly into a product of three factors: the confounding, the discrepancy, and the variation. This identity, introduced by Xiaoli Meng of Harvard University, arises in a number of settings, including deterministic sampling, randomized sampling, and Bayesian cubature. It provides insight that traditional analyses do not. This talk gives an overview of the trio identity in various settings, and illustrates it be means of a few examples. The first example is understanding the value of a web survey.
Friday, November 4, 2016 at 3:30pm to 4:30pm
Levan Center, 201
2322 N Kenmore Ave
No recent activity