|
Activity Number:
|
303
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 11, 2004 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #302104 |
|
Title:
|
Statistical Computing in Neutrino Signal Detection
|
|
Author(s):
|
Fan Lu*+ and Grace Wahba and Gary Hill and Paolo Desiati
|
|
Companies:
|
University of Wisconsin, Madison and University of Wisconsin, Madison and University of Wisconsin, Madison and University of Wisconsin, Madison
|
|
Address:
|
1210 W. Dayton St., Madison, WI, 53706,
|
|
Keywords:
|
penalized log-likelihood ; nonstandard support vector machine ; Kullback-Leiber distance ; logit function ; optimization ; neutrino signal
|
|
Abstract:
|
Physicists are trying to use the giant device called AMANDA (Antarctic Muon and Neutrino Detector Array) buried deep in the Antarctic ice cap to detect certain neutrino signals within comparatively overwhelming background noise. Distributions of signal and background are generated by an importance sampling procedure that generates events described by multiple feature variables, computable from AMANDA data. Each event is labeled with an importance sampling weight for signal and one for background. The task is to find the most powerful decision boundary at certain significance level to distinguish signal neutrino from background neutrino. Because of the curse of dimensionality, usual Monte Carlo methods are not practical. We first propose a modified penalized log-likelihood approach to estimate the logit function in this scenario, which involves two major optimization steps and the use of KL (Kullback-Leibler) distance criterion for model tuning. Then we may compare this approach with a nonstandard SVM (support vector machine) approach. Simulation studies are presented for both approaches before we show the final results for the neutrino data.
|
