Data-Efficient Quickest Change Detection

Abstract

In the classical problem of quickest change detection, a decision maker observes a sequence of random variables. At some point in time, the distribution of the random variables changes abruptly. The objective is to detect this change in distribution with minimum possible delay, subject to a constraint on the false alarm rate. In many applications of quickest change detection, e.g., where the changes are infrequent, it is of interest to control the cost of observations or the cost of data acquired before the change point. To this end, in this survey paper, data-efficient versions of the classical quickest change detection problems are studied, and a summary of existing results on these problems is provided. Some extensions to distributed sensor networks are also discussed.

Quickest intrusion detection in mobile active user authentication

“…Quickest Change Detection (QCD) is a branch of statistical signal processing that thrives to detect the change point of statistical properties of a random process [14], [15], [16]. The objective of QCD is to produce algorithms that detect the change with a minimal delay (ADD) while adhering to false alarm rate constraints (PFD).…”

“…Then, average percentage of observations (APO) obtained prior to the change point can be quantified as AP O = E 1 S S n=1 Mn , where E denotes the Expected value.In a non-Bayesian setting, due to the absence of a priori distribution on the change point, a different quantity should be used to quantify the number of observations used for decision making. Work in[15],[16], proposes Prechange Duty Cycle (PDC) as P DC = lim sup n En n−1 k=1 M k |τ ≥ n for this purpose. It should be noted that both PDC and APO are similar quantities.…”