Success in nuclear forensics search is a critical component to fighting terrorist activity and preventing disastrous individual terrorist nuclear attacks. The UC Berkeley Nuclear Forensic Search Project takes a computer science algorithmic approach (as a special directed graph matching problem) to address nuclear forensics search, essentially recasting nuclear forensics discovery as a digital library search problem. A simultaneous aim is to encourage other computer scientists to work on nuclear forensics search.
This talk will describe our project, which has been funded by the National Science Foundation and the DHS Domestic Nuclear Detection Office's Academic Research Initiative. After historical background on nuclear forensics, we will focus on three approaches to identification of sources of interdicted nuclear material -- matching based upon properties of isotopes and isotope ratio measurements, exclusion based upon machine learning to identify nuclear spent fuel by reactor type or uranium ores by geologically specific element composition, and capturing the logic of the forensic process by which a human nuclear forensic expert would engage the attribution challenge. We will describe the results of our preliminary search experiments using the OECD-Nuclear Energy Agency's SFCOMPO Spent Fuel Database, which will be presented in mid-September at the German Informatik (LWA) conference at the University of Dortmund