Asymptotically Optimal Sampling Policy for Quickest Change Detection With Observation-Switching Cost

We consider the problem of quickest change detection (QCD) in a signal where its observations are obtained using set actions, and switching from one action to another comes with cost. The objective is design stopping rule consisting sampling policy determine sequence actions used observe time quickly detect for change, subject constraint on average observation-switching propose an open-loop finite window size generalized likelihood ratio (GLR) Cumulative Sum (CuSum) QCD problem. show that GLR CuSum asymptotically optimal properly designed formulate this as quadratic programming prove it sufficient policies not more than when designing several algorithms solve optimization theoretical guarantees. Finally, we apply our approach partially observed graph empirically demonstrate performance proposed times.