Change-point Detection, Then and Now
About the Speaker
David Siegmund, who holds the John D. and Sigrid Banks Chair at Stanford University, Stanford, CA, is a statistician who is comfortable in both the airy heights of theory and the practicalities of real-world applications. He works at the interface between probability and statistics, applying the tools he develops to topics as diverse as the design of medical clinical trials and mapping the locations of genes that are involved in specific physiological traits. His work has earned him several awards, including a Guggenheim Fellowship in 1974, the Humboldt Prize in 1980, and membership in the American Academy of Arts and Sciences in 1994. In 2002 he was elected to the National Academy of Sciences.
Change-point detection had its origins almost sixty years ago in the work of Page (UK), Shiryayev (USSR), and Lorden (US), who focused on sequential detection of a change-point in an observed process. The process was typically a model for the quality of the output of a production process that began in an acceptable state, and the change-point indicated a deterioration in quality that must be detected and corrected. Recently, motivation from a broad range of applications has lead to a variety of different problems, involving both retrospective fixed sample and online detection of change-points. I will review this history with emphasis on contemporary applications in biology and common features of likelihood based approaches.