From the Spring 2008 issue of Threads
Jeff Philips got started on research early. While an undergrad at Rice University, he worked in Lydia Kavraki’s lab, publishing papers on physical simulation and robotic motion planning, which appeared in IEEE Conference on Robotics and Automation, the main conference in robotics. After graduating with a BS in Computer Science and a BA in Mathematics, he joined the PhD program at Duke in Fall 2003, with the goal of working on something more mathematical but still geometric. Phillips’s undergraduate research experience helped him win a James B. Duke graduate fellowship and a National Science Foundation graduate research fellowship.
Soon after arriving at Duke, Phillips started working with Pankaj Agarwal. Motivated by problems in protein interaction as part of the Biogeometry project, Phillips became interested in shape-analysis problems. He has contributed on both theoretical and practical levels. His work has provided some of the first algorithms with formal theoretical guarantees for matching point sets allowing transformations while using the ubiquitous root mean squared distance. He also developed a practical method for the popular ICP point matching algorithm to handle outliers. In collaboration with Johannes Rudolph, a biochemist, Phillips spearheaded a project to detect geometric motifs on the interface between two bound proteins. These motifs have potential to identify residues critical to the binding process. This piece of work has been integrated into the Biogeometry’s Protein-Protein Interface Surface MAPS software.
As the Biogeometry project wound down, Phillips’s focus shifted more toward the interface between computational geometry problems and statistics. Stemming from a summer as a visitor at AT&T Shannon Labs, Phillips became interested in a statistical anomaly detection technique for spatial data. The algorithm he and his collaborators developed and implemented was the first for this class of problems to have a guaranteed bound on its approximation, and it also performs comparable or better to known heuristic techniques in practice.
More recently, Phillips has been extending these techniques and looking at geometry problems that lie at the foundation of many spatial statistical questions. He is also applying his technique to develop better uncertainty models for geometric problems.
Bringing mathematical rigor to geometric problems that arise in many applied fields and developing simple, efficient algorithms for them have been a recurring theme of Phillips’s research. What sets Phillips apart from many of his peers is his ability to work independently, to collaborate with a broad spectrum of researchers, and to juggle between many problems at the same time. Over his brief research career, Phillips has published in fields ranging from robotics to bioinformatics, from algorithms to modeling, and from geometry to databases and data mining.
Besides being an outstanding researcher, Phillips has also been a devoted citizen of the Department. The Department has benefited tremendously from his leadership and mentoring skills. New students flock to him for advice, and he is always very generous with his time. Phillips thrice chaired the graduate student recruitment, organized the design of department t-shirts, curated two seminars, and served as the graduate student liaison to the faculty. He won the Outstanding Department Service Award in 2006.