Back to search

Institutes outside Greece

Sequential anomaly detection under sampling constraints

Abstract

dc:description

The problem of sequential anomaly detection is considered, where multiple data sources are monitoredin real-time, and the goal is to identify the ones that exhibit outlying statistical behavior,when it is not possible to sample all sources at all times. A detection scheme in this context requiresspecifying not only when to stop sampling and which sources to identify as anomalous uponstopping but also which sources to sample at each time instance until stopping. A novel formulationfor this problem is proposed, in which the number of anomalous sources is not necessarilyknown in advance and the number of sampled sources per time instance is not necessarily fixed.Instead, an arbitrary lower bound and an arbitrary upper bound are assumed on the number ofanomalous sources, and the fraction of the expected number of samples over the expected timeuntil stopping is required not to exceed an arbitrary, user-specified level. In addition to this samplingconstraint, the probabilities of at least one false alarm and at least one missed detection arecontrolled below user-specified tolerance levels. A general criterion is established for a policy toachieve the minimum expected time until stopping to a first-order asymptotic approximation asthe two familywise error rates go to zero. Moreover, asymptotic optimality is established for twofamilies of sampling policies. In the first family, named as probabilistic family, each sampling rulechooses each source at each time instance with a probability that depends on past observationsonly through the current estimate of the subset of anomalous sources. In the second family, namedas ordering family, each sampling rule decides the sources to sample based on the ordering of thelikelihood ratio statistics, where randomization may be applied only for the treatment of the decimalpart of the sampling size. The detection policies can also be adapted to treat generalized errormetrics. We examine two error metrics: in the first one, we control the probability of a prescribednumber of misclassification errors, and in the second one, both the probabilities of a number offalse alarms and a number of missed detections. We also study the sequential anomaly detectionproblem from the perspective of the composite hypotheses setting. In the composite hypothesessetting, the samples collected from each source follow the model of a distribution parameterizedby a parameter that belongs to one of two disjoint subsets of a compact Euclidean space. The twocandidate subsets are not necessarily the same for each source. If the parameter belongs to the firstsubset, we characterize the source as outlying; otherwise, we characterize it as regular. The samiiples received from different sources are independent of each other. In all frameworks, we providesimulations that compare the performance of the sampling rules from both families for small fixedthresholds on the error probabilities.

Degree

thesis:*
Grantor dc:publisher
Institutes outside Greece
Year dc:date
2023

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Tsopelakos, Aristomenis
  • Τσοπελάκος, Αριστομένης

Subjects

dc:subject × 8

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/60766

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Tsopelakos, Aristomenis; Τσοπελάκος, Αριστομένης. Sequential anomaly detection under sampling constraints. Institutes outside Greece, 2023. http://hdl.handle.net/10442/hedi/60766