The problem of detecting changes in stochastic systems, often referred to as sequential change detection or quickest change detection, arises in various branches of science and engineering, including the various signal processing applications considered in this research program. In all these applications, an anomaly in the environment changes in some way the distribution of the sequentially acquired observations. The goal is to detect the change and raise an alarm as soon as possible, so that any necessary action can be taken in time, while controlling the rate of false alarms below an acceptable level.
While the quickest change detection problem has been actively studied since early 1950s, there are many open challenges in this field that are of theoretical as well as practical interest. Specifically, it is important to design sequential change detection rules that are efficient (optimal or nearly optimal) when observations are acquired at multiple streams, obtained by sensor, cyber and social network, and the change may have a sparse structure. Moreover, it is important to move beyond the traditional i.i.d. (independent and identically distributed) assumption for the observations, allowing for different levels of uncertainty in the post-change regime and dependence in the observations. In this project, our focus will be on a number of challenging open problems in these directions. Our ambition is to develop the next generation of quickest detection methods, as well as their underlying theory.