Algorithm to help soldiers identify safe, enemy shipments
In another project supported by the U.S. Department of Defense, Holder is working to analyze relationships in war zones to try to help the military better identify its friends and enemies.
Much intelligence collection is done by word of mouth, he said. Officers conduct interviews daily, take notes and try to piece together a social network of who knows whom, or who might be working for whom. Based on the relationships, officers identify those who may be a friend or an enemy.
Holder is developing computer programs that will help analyze the data to uncover anomalies. With the help of the algorithm, soldiers could enter data, keep track of and share relationship information through their smart phones or personal digital assistants (PDA).
The algorithm simply helps automate information gathering, so soldiers can more quickly and efficiently understand the relationships around them. And it could help soldiers better determine what piece of information might be missing or who they should interview next.
Holder’s team hopes to begin deploying the program on soldier training missions within three years.
PULLMAN – Sue knows a drug smuggler, and the drug smuggler visits Canada, New Jersey and Jamaica. John, who has visited Connecticut, New Jersey and Jamaica, knows Sue. Bob and Bill, who know Sue, also have been to Canada. Who is the drug smuggler?
Trying to stop a smuggler from bringing drugs into the country – or other such illegal activity – might be thought of as a tricky logic puzzle, said Larry Holder, a professor in the School of Electrical Engineering and Computer Science – except that instead of trying to solve who is the drug smuggler among a few like Sue, Bob, Bill or John, there are a million cargo shipments that come into the U.S. every month from around the world.
Holder recently received a three-year grant from the U.S. Department of Homeland Security to better understand patterns in cargo data to help the agency better determine which shipments might be suspect.
Holder conducts research on graph-based learning and has developed algorithms to better understand relationships in data.
The cargo at U.S. ports is already regularly screened, Holder said, but he is working to help inspectors know better which cargo, in particular, should be targeted for further screening and inspection. So his team developed an algorithm that can look at the many attributes of each piece of cargo.
The researchers hope the tool might be tested in real ports in three years.
The computer program looks at a variety of data, such as who owns the cargo, what other companies it might be connected with, where the shipment originated and which ports it has traveled through. A graphical network is developed for each piece of cargo, and then the computer program looks for anomalies in the vast amounts of data.
For instance, to test their algorithm, the researchers simulated a drug shipment that actually occurred in 2000. The researchers knew that the cargo ship had made an unscheduled stop in the Bahamas and some financial information had been left off the shipping form. Out of a group of shipments, the algorithm developed by Holder was able to flag the unusual pattern of the cargo that contained the drugs.
With the DHS grant, the researchers are working to improve the algorithm so it reliably can be used in ports. It must be able to detect anomalies that are helpful without overburdening the system with too many false positive results.
Not delaying cargo
The vast majority of shippers presumably are not participating in criminal behavior, so the computer program has to be able to pick out the very small percentage of those who are. Furthermore, the data has to be analyzed and the algorithm has to work efficiently and quickly, so that the cargo making its way through ports is not delayed unnecessarily.
“The data could be analyzed for years, but we have to analyze it before the next shipment comes in,’’ Holder said.
High accuracy results
The researchers are continuing to test the computer program by injecting a piece of cargo with known anomalies into a sea of data and letting the program find the anomaly.
“So far, we’ve been able to find it every time,’’ Holder said.