Using Data Competitions to Crowdsource Innovative Solutions to Urban Radiation Detection Problems

SPEAKER: CHRISTINE ANDERSON-COOK DATE/TIME: MON, 03/02/2020 – 4:00PM TO 5:00PM LOCATION: 3105 ETCHEVERRY HALL Spring 2020 Colloquium Series

Abstract:

In 2017 and 2018, NA-22 sponsored a project to host data competitions to solicit innovative solutions for urban radiation detection problems. A team from Los Alamos, Oak Ridge, and Berkeley National Laboratories fielded and hosted two competitions. The first was restricted to those with a government affiliation, while the second was hosted on Topcoder and was open to an international field of competitors. The competitors were asked to detect, identify and locate 6 different radioactive sources from simulated data that modeled a detector being driven down an urban street. A total of 87 competitors created over 2500 submissions across the two competitions. We will present a new methodology for creating the training and test data sets for the competition that thoroughly explores the diverse problems space considered. In addition, an extended post-competition analysis is able to highlight the strengths and weaknesses of the winning algorithms and make detailed comparisons between competitors. A different version of the problem may want to emphasize aspects of the detect, identify and locate the problem for runs with sources and false positives for runs with no sources. Hence, multiple criterion optimization with Pareto fronts allows identification of top solutions for different combinations of the study objectives. Results from the two competitions will be presented and the solution with their improvement of currently used methods compared. 

About the Speaker:

Christine Anderson-Cook has been a Research Scientist in the Statistical Sciences Group at Los Alamos National Laboratory since 2004. She currently leads projects in the areas of sequential design of experiments, non-proliferation and effective hosting of data competitions. Before joining LANL, she was a faculty member in the Department of Statistics at Virginia Tech for 8 years. Her research areas include response surface methodology, design of experiments, reliability, multiple criterion optimization and graphical methods. She has authored more than 200 articles in statistics and quality peer-reviewed journals, and has been a long-time contributor to the Quality Progress Statistics Spotlight column. She has co-authored a popular book on Response Surface Methodology with Raymond Myers and Douglas Montgomery. She has served on the editorial boards of Technometrics, Journal of Quality Technology, Quality Engineering and Quality and Reliability Engineering International. She is an elected fellow of the American Statistical Association and the American Society for Quality. She is the recipient of the ASQ 2018 Shewhart Medal, the ASQ Statistics Division 2012 William G. Hunter Award, and a two-time recipient of the ASQ Shewell Award. In 2011 she received the 26th Annual Governor’s Award for Outstanding New Mexico Women.