Student's AF-funded research to work complex problems Published Feb. 6, 2009 By Maria Callier Air Force Office of Scientific Research ARLINGTON, Va. (AFNS) -- A doctoral student at the Massachusetts Institute of Technology is designing Air Force Office of Scientific Research-supported computer models that can describe the pattern of complicated scenarios ranging from finding oil underground to pinpointing suspicious behaviors in urban settings. Emily Fox's research offers promise for military applications because of its applications for target intent or changes in target behavior that might signal danger to military or civilians. "The defense element of this work opens up possibilities of learning coordinated behaviors of multiple targets," said Dr. Alan S. Willsky, Ms. Fox's Ph.D. thesis supervisor at MIT. Dr. Willsky suggested Ms. Fox try a challenging new technical project: "the examination of dynamically evolving phenomena and the discovery and extraction of coherent behavior with minimal prior information." "She has succeeded in exceptional fashion and has now developed a suite of methods that have very broad applicability," he said. For the context of oil exploration, Ms. Fox differs somewhat in her approach because she analyzes geophysical data to reveal environments where oil may be found beneath the surface. "At an intellectual level, these methods are a major advance in statistical signal and data processing, machine learning, dynamic estimation, tracking and prediction. There are some deep and conceptually very important tools that Fox has used to develop these algorithms," Dr. Willsky said. Dynamic systems modeling that preceded Ms. Fox's research usually used background information, typically in the form of a single or a specified set of behavior models as well as prior knowledge on how switches among these behaviors were likely to occur. Potential future applications include constructing models that will reveal patterns of motion of groups of objects -- for example, routes taken by groups of vehicles from particular origins to particular destinations. These extracted models might then be used to determine irregular behavior inconsistent with extracted patterns of normal behavior. "The promise of this modeling formalism is to provide a way in which the available data can very naturally speak for itself and provide model complexity without being guided," said Dr. Willsky. A natural extension of her work is the development of patterns of behavior of entire groups of people that are exhibiting the same behaviors. "We have just scratched the surface of this interesting area, but Fox has certainly created a very powerful foundation," said Dr. Willsky. Comment on this story (comments may be published on Air Force Link) View the comments/letters page