Alexander Tartakovsky
Title: Quickest Changepoint Detection: Recent Advances and Open Problems
Abstract: We will discuss the quickest (sequential) changepoint problem when one is interested in detecting a change as soon as possible after its occurrence while maintaining a false alarm rate at a given level. A classical setting deals with the iid case where observations are iid in the pre-change mode with one distribution and iid post-change with another distribution. We will consider a general non-iid setup, which covers Markov, hidden Markov, and state-space models as particular cases. We will also address certain challenging applications such as network security (rapid detection of attacks in computer networks) and surveillance (detection and tracking of low-observable targets). Certain open problems will be discussed. Part of this research has been triggered by the joint work with Bal in 1993-94.