This talk was delivered to the University of Texas Operations Research and Industrial Engineering seminar in February 2012.  The abstract is as follows.

In many intelligence agencies, the screening of data into usable information ready for analysis poses a significant bottleneck. Typically, much more data is available than what can be screened in the allotted time. We call the staff who screen raw data into usable information "collectors."

We formulate the problem faced by an intelligence collector--selecting which data to screen--as an exploration-exploitation problem: the collector has to choose between exploring for new sources of relevant information and exploiting known sources.

To address the exploration-exploitation problem, we develop a mathematical model of the collector’s knowledge and examine algorithms that allow the collector to maximize the discovery of relevant data given a time limit. We derive insights on the performance of different algorithms using a simulated case study. 

Goto feature

Humanitarian assistance is of growing importance to the United States and the Department of Defense’s strategic objectives. Thus, United States combatant commands increasingly rely on humanitarian assistance cargo transportation programs to deliver material to people in need in their areas of responsibility. This report analyzes the options available to these commands in seeking humanitarian assistance cargo transportation. The report offers a description of current operations, with a specific focus on the European area of responsibility, where humanitarian assistance cargo transportation programs have had limited activity.

Goto feature

The U.S. Centers for Disease Control have asked each of the U.S. states to maintain an influenza surveillance network. The network consists of a number of primary healthcare providers who report weekly to the state on the number of patients that visited their office, and the number of those patients that exhibited influenza-like-illness symptoms. Each state has recruited its own providers into its surveillence network.

In this talk, we discuss how to create a well-performing influenza surveillence network. There are thousands of possible providers which the state could recruit into the surveillance network. Which of them should the state recruit? How should those providers be spread out geographically across the state? And can new technologies like Google Flu Trends replace or augment traditional surveillance networks?

Goto feature

Different radioactive materials and their associated decay chains produce different radiation energy spectra.  Naturally, this has lead to the study of identifying a radioactive material based on its energy spectrum, called radiation spectroscopy.  One of the central problems in radiation spectroscopy is the resolution of the radiation detector.  As radiation hits the detector, the noise from the detection equipment and the random process of radiation transportation smears out the radiation spectrum of the radioactive material, making wide energy peaks instead of sharp ones.

Seeded Localized Averaging (SLA) is a real-time spectrum aquisition method that allows us to sharpen the peaks in aquired energy spectra.  SLA was patented by my co-author, Valentin Jordanov, in 2007.  In this talk, presented in ANIMMA 2011, for the first time, we prove some basic theoretical properties of the SLA transformation.  We also describe the transformation and show some practical examples.

Goto feature

My talk from the Epidemics² conference in Athens describes our recent application of bandit-based search for computing antiviral distributions for the U.S. National Antiviral Stockpile.

At the beginning of the summer in 2009, the U.S. government learned of the swine-origin influenza outbreak.  Immediately, the government released 11 million courses of antivirals from the national stockpile for immediate use by the states.  The states themselves had purchased some antivirals, totaling 35 million courses available for immediate use.  Additionally, the U.S. government had 50 million courses in reserve in the national stockpile.  We address the question of how and when antivirals from U.S. National Antiviral Stockpile should be released to best control an influenza outbreak.  We do this quantitatively, using a national scale disease model and recent simulation optimization techniques.

Goto feature