Please note: This PhD seminar will be given online.
Aida Sheshbolouki, PhD candidate
David R. Cheriton School of Computer Science
Supervisor: Professor M. Tamer Özsu
Concept Drift (CD) occurs when a change in insufficient, unknown, or unobservable information (a hidden context) about data induces changes in known and/or observable information (a target context) about data. CD is a natural phenomenon in streaming data due to the non-stationary setting. CD understanding, detection, and adaptation is vital for effective and efficient analysis/analytics over streaming data as reliable output depends on fresh input. The research in CD is mostly focused on cases where data instances are assumed to be independent, and the target concept is defined based on labelled data instances with respect to a supervised down stream task.
Taking this literature as the stepping stones, this talk starts with a discussion on the essence, challenges, and potential impacts of studying CD in case of transient, interconnected, and sequential data instances forming a streaming graph which serves as the input to any analytic task. The talk will then continue with introducing a proposed framework for understanding and detection of CD in streaming graphs and experimental findings.