Abstract:

Every year millions of containers reach US ports, and only a small part of them is inspected thoroughly. There are heightened concerns recently regarding illegal nuclear material hidden potentially in some of these containers. In the inspection process a number of different imaging technologies are used, each providing only partial information about the content of the containers. We can safely assume that even if we had known all possible sensor readings, it would be impossible to tell exactly if a container contains some dangerous material, or not. Moreover, the application of these sensors is expensive, and it is not feasible to assume that we can apply each of the sensors to each of the containers arriving in our ports. We present here a tractable mathematical model for this situation to maximize detection rate while keeping expected cost below given budget. The model is based on a polyhedral characterization of decision trees.

Joint work with Liliya Fedzhora (RUTCOR), Paul B. Kantor (SCILS, Rutgers), Kevin Saeger (LANL), and Philip Stroud (LANL).