BEGIN:VCALENDAR VERSION:2.0 PRODID:-//TYPO3/NONSGML Calendarize//EN BEGIN:VEVENT UID:calendarize-concept-drift-detection-in-unlabeled-data-streams DTSTAMP:20260413T114427Z DTSTART:20260427T080000Z SUMMARY:Concept Drift Detection in Unlabeled Data Streams DESCRIPTION:Am Montag\, den 27. April 2026\, um 10:00 Uhr hältDaniel Luka ts Universität Oldenburgim Rahmen seiner beabsichtigten Dissertation eine n Vortrag mit dem TitelConcept Drift Detection in Unlabeled Data StreamsDe r Vortrag findet online statt: https://studconf.uol.de/rooms/mpz-niv-bz5-4 0a/join Der Vortrag erfolgt in englischer Sprache.\nAbstract:In long-runn ing data streams\, an issue called concept drift may arise. Concept drift denotes changes in the feature distribution P(X) or the posterior distribu tion P(Y | X)\, such that the performance of classifiers employed on the d ata stream may deteriorate severely. Besides detecting concept drifts for the adaptation of classifiers\, concept drifts themselves may be worth det ecting\, e.g.\, in marine domains where changes such as ocean fronts or al gal blooms can be interpreted as concept drifts which merit further action .Although various concept drift detectors have been suggested in the liter ature\, most detectors are supervised: They require and assume the availab ility of the true label immediately after predicting it. This assumption i s hard to justify\, since labeling a potentially infinite data stream in r eal time can become prohibitively expensive.While a number of unsupervised approaches to concept drift detection exist\, few have been demonstrated on real-world applications. Instead\, these methods are commonly evaluated on artificially created problems without ground truth information about c oncept drift. Thus\, proxy metrics are commonly used in the literature. Ho wever\, these proxies are flawed\, as they often reward frequent false pos itives more than correct detection.Based on this interpretation of the sta te of the art\, this work discusses the following research questions: How can unsupervised concept drift detectors detect concept drift on continuou s features? How can the predictive quality and the timeliness of concept d rift detectors be evaluated in an unbiased fashion?To this end\, existing metrics such as the F1 Score are adapted for use in concept drift detectio n. Then a system encompassing an unsupervised concept drift detector and o ptional components to address common real-world properties such as seasona l data\, trend and noisy features is evaluated. These research questions a re discussed on synthetic data streams and in the context of oceanographic data streams with a case study on the detection of algal blooms in the Ba ltic Sea.Betreuer: Prof. Dr.-Ing. Axel Hahn X-ALT-DESC;FMTTYPE=text/html:

Am Montag\, den 27. Ap ril 2026\, um 10:00 Uhr hält
Daniel Lukats Universität Oldenburg
im Rahmen seiner beabsichtigten Dissertation einen Vortrag mit dem Tite l
Concept Drift Detection in Unlabeled Data Streams
Der Vortrag findet online statt: https://studconf.uol.de/rooms/mpz-niv-bz5-40a/join \;
Der Vortrag erfolgt in englischer Sprache.

\n

Abstract:
In long-running data streams\, an issue called concept drift may arise. Concept drift denotes changes in the feature distribution P(X) or the posterior distribution P(Y | X)\, such that the performance of classif iers employed on the data stream may deteriorate severely. Besides detecti ng concept drifts for the adaptation of classifiers\, concept drifts thems elves may be worth detecting\, e.g.\, in marine domains where changes such as ocean fronts or algal blooms can be interpreted as concept drifts whic h merit further action.
Although various concept drift detectors have been suggested in the literature\, most detectors are supervised: They re quire and assume the availability of the true label immediately after pred icting it. This assumption is hard to justify\, since labeling a potential ly infinite data stream in real time can become prohibitively expensive.While a number of unsupervised approaches to concept drift detection e xist\, few have been demonstrated on real-world applications. Instead\, th ese methods are commonly evaluated on artificially created problems withou t ground truth information about concept drift. Thus\, proxy metrics are c ommonly used in the literature. However\, these proxies are flawed\, as th ey often reward frequent false positives more than correct detection.
Based on this interpretation of the state of the art\, this work discusse s the following research questions: How can unsupervised concept drift det ectors detect concept drift on continuous features? How can the predictive quality and the timeliness of concept drift detectors be evaluated in an unbiased fashion?
To this end\, existing metrics such as the F1 Score are adapted for use in concept drift detection. Then a system encompassin g an unsupervised concept drift detector and optional components to addres s common real-world properties such as seasonal data\, trend and noisy fea tures is evaluated. These research questions are discussed on synthetic da ta streams and in the context of oceanographic data streams with a case st udy on the detection of algal blooms in the Baltic Sea.
Betre uer: Prof. Dr.-Ing. Axel Hahn

LOCATION: END:VEVENT BEGIN:VEVENT UID:calendarize-departmentrat DTSTAMP:20260316T162603Z DTSTART:20260429T110000Z DTEND:20260429T120000Z SUMMARY:Departmentrat DESCRIPTION: X-ALT-DESC;FMTTYPE=text/html: LOCATION:Meeting Raum in StudIP-Veranstaltung 2.01.888 DPR Informatik (htt ps://elearning.uni-oldenburg.de/dispatch.php/course/details?sem_id=e018045 1c9ee9edb1c3ba4fd34e5fbc6&sso=cas&cancel_login=1&ticket=ST-4524636-MY3KWgB yBQtaitZAKTcwvr3fwDUcas05) END:VEVENT END:VCALENDAR