This research will explore a broad class of recursive robust estimation procedures. Recursive procedures are important when it is too expensive to store all of the previous data and when real-time estimates are required. Update procedures can provide good approximate estimates at lower costs. The particular problem to be investigated relates to abrupt changes in parameters of static or dynamic models that arise in many areas. This type of problem is called jump, failure, or fault detection in engineering and in quality control, and as detection of model instability or change point analysis in statistics and econometrics. Much of the past work assumes an underlying Gaussian distribution. The principal investigator will explore more resistent recursive procedures.
